perf(video): 优化视频处理性能监控和音频播放

- 添加视频处理性能计时和统计功能
- 实现帧处理时间监控和慢帧警告
- 添加音频文件静音修剪功能
- 优化Windows平台音频播放实现
- 调整默认日志输出频率减少冗余信息
- 修复MediaPipe GPU委托在Windows上的兼容性问题

.gitignore

@@ -1,3 +1,6 @@
 .venv/
 .idea/
 __pycache__/
+logs/
+
+resources/

README.md

@@ -0,0 +1,10 @@
+# PoseFit Server
+
+Real-time exercise pose detection and coaching via WebRTC.
+
+## Quick Start
+
+```
+pip install -r requirements.txt
+python run.py
+``

app/audio/generate.py

@@ -0,0 +1,227 @@
+# app/audio/generate.py
+from __future__ import annotations
+
+import platform
+import shutil
+import subprocess
+import wave
+from pathlib import Path
+
+from loguru import logger
+
+
+def generate_rep_audio_files(
+    *,
+    max_count: int,
+    rate: int,
+    output_dir: Path,
+    overwrite: bool = False,
+    trim_leading_silence: bool = True,
+    trim_silence_threshold: int = 500,
+    trim_silence_padding_ms: int = 20,
+) -> None:
+    """
+    确保 0~max_count 的运动次数语音 wav 文件存在。
+
+    默认生成到：
+
+        resources/audio/reps/0.aiff  # macOS
+        resources/audio/reps/0.wav   # Windows / Linux
+        ...
+        resources/audio/reps/200.aiff 或 200.wav
+
+    服务启动时调用一次即可。
+    """
+    output_dir.mkdir(parents=True, exist_ok=True)
+    system = platform.system().lower()
+    suffix = ".aiff" if system == "darwin" else ".wav"
+
+    missing_counts = [
+        count
+        for count in range(0, max_count + 1)
+        if overwrite or not _audio_path(output_dir, count, suffix=suffix).exists()
+    ]
+
+    if not missing_counts:
+        logger.info("Rep audio files already prepared: {}", output_dir)
+    else:
+        logger.info(
+            "Preparing rep audio files, system={}, count={}, output_dir={}",
+            system,
+            len(missing_counts),
+            output_dir,
+        )
+
+        if system == "darwin":
+            _generate_with_macos_say(
+                counts=missing_counts,
+                output_dir=output_dir,
+                rate=rate,
+            )
+        else:
+            _generate_with_pyttsx3(
+                counts=missing_counts,
+                output_dir=output_dir,
+                rate=rate,
+            )
+
+        logger.info("Rep audio files prepared: {}", output_dir)
+
+    if trim_leading_silence and suffix == ".wav":
+        _trim_leading_silence_files(
+            counts=list(range(0, max_count + 1)),
+            output_dir=output_dir,
+            suffix=suffix,
+            threshold=trim_silence_threshold,
+            padding_ms=trim_silence_padding_ms,
+        )
+
+
+def _generate_with_macos_say(
+    *,
+    counts: list[int],
+    output_dir: Path,
+    rate: int,
+) -> None:
+    """macOS 使用 say 命令生成 wav。"""
+    if platform.system().lower() != "darwin":
+        raise RuntimeError("say command is only available on macOS")
+
+    if shutil.which("say") is None:
+        raise RuntimeError("macOS say command not found")
+
+    for count in counts:
+        audio_file = _audio_path(output_dir, count, suffix=".aiff")
+
+        try:
+            subprocess.run(
+                [
+                    "say",
+                    "-r",
+                    str(rate),
+                    "--file-format=AIFF",
+                    "-o",
+                    str(audio_file),
+                    str(count),
+                ],
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.PIPE,
+                text=True,
+                check=True,
+            )
+        except subprocess.CalledProcessError as exc:
+            message = exc.stderr.strip() or f"exit status {exc.returncode}"
+            raise RuntimeError(f"Failed to generate {audio_file}: {message}") from exc
+
+
+def _generate_with_pyttsx3(
+    *,
+    counts: list[int],
+    output_dir: Path,
+    rate: int,
+) -> None:
+    """Windows / Linux 使用 pyttsx3 生成 wav。"""
+    try:
+        import pyttsx3
+    except Exception as exc:
+        raise RuntimeError(f"pyttsx3 unavailable: {exc}") from exc
+
+    engine = pyttsx3.init()
+    engine.setProperty("rate", rate)
+    engine.setProperty("volume", 1.0)
+
+    for count in counts:
+        audio_file = _audio_path(output_dir, count, suffix=".wav")
+        engine.save_to_file(str(count), str(audio_file))
+
+    engine.runAndWait()
+
+
+def _audio_path(output_dir: Path, count: int, *, suffix: str) -> Path:
+    return output_dir / f"{count}{suffix}"
+
+
+def _trim_leading_silence_files(
+    *,
+    counts: list[int],
+    output_dir: Path,
+    suffix: str,
+    threshold: int,
+    padding_ms: int,
+) -> None:
+    trimmed = 0
+    total_removed_ms = 0.0
+
+    for count in counts:
+        audio_file = _audio_path(output_dir, count, suffix=suffix)
+        if not audio_file.exists():
+            continue
+        removed_ms = _trim_leading_silence(audio_file, threshold=threshold, padding_ms=padding_ms)
+        if removed_ms > 0:
+            trimmed += 1
+            total_removed_ms += removed_ms
+
+    logger.info(
+        "Rep audio leading silence trim complete: files_trimmed={}, total_removed_ms={:.1f}, threshold={}, padding_ms={}",
+        trimmed,
+        total_removed_ms,
+        threshold,
+        padding_ms,
+    )
+
+
+def _trim_leading_silence(audio_file: Path, *, threshold: int, padding_ms: int) -> float:
+    with wave.open(str(audio_file), "rb") as reader:
+        params = reader.getparams()
+        frames = reader.readframes(params.nframes)
+
+    frame_size = params.sampwidth * params.nchannels
+    if params.nframes <= 0 or frame_size <= 0:
+        return 0.0
+
+    chunk_frames = max(1, params.framerate // 100)
+    leading_frames = 0
+    offset = 0
+    chunk_size = chunk_frames * frame_size
+
+    while offset < len(frames):
+        chunk = frames[offset : offset + chunk_size]
+        if _pcm_rms(chunk, params.sampwidth) > threshold:
+            break
+        chunk_frame_count = len(chunk) // frame_size
+        leading_frames += chunk_frame_count
+        offset += chunk_size
+
+    padding_frames = int(params.framerate * max(0, padding_ms) / 1000)
+    remove_frames = max(0, leading_frames - padding_frames)
+    if remove_frames <= 0:
+        return 0.0
+
+    start = min(len(frames), remove_frames * frame_size)
+    trimmed_frames = frames[start:]
+    if not trimmed_frames:
+        return 0.0
+
+    with wave.open(str(audio_file), "wb") as writer:
+        writer.setparams(params)
+        writer.writeframes(trimmed_frames)
+
+    return remove_frames / params.framerate * 1000
+
+
+def _pcm_rms(chunk: bytes, sample_width: int) -> float:
+    if not chunk:
+        return 0.0
+
+    if sample_width == 2:
+        sample_count = len(chunk) // 2
+        if sample_count == 0:
+            return 0.0
+        total = 0
+        for i in range(0, sample_count * 2, 2):
+            sample = int.from_bytes(chunk[i : i + 2], "little", signed=True)
+            total += sample * sample
+        return (total / sample_count) ** 0.5
+
+    peak = max(abs(byte - 128) for byte in chunk)
+    return float(peak)

app/audio/rep_announcer.py

@@ -0,0 +1,178 @@
+from __future__ import annotations
+
+import queue
+import shutil
+import subprocess
+import sys
+import threading
+import time
+from pathlib import Path
+
+from loguru import logger
+
+
+class RepAnnouncer:
+    """运动次数语音播报器：读取预生成的音频文件直接播放"""
+
+    def __init__(
+        self,
+        *,
+        enabled: bool = True,
+        max_count: int = 200,
+        audio_dir: str | Path = "resources/audio/reps",
+    ) -> None:
+        self.enabled = enabled
+        self.max_count = max_count
+        self.audio_dir = Path(audio_dir)
+
+        self._queue: queue.Queue[tuple[int, float] | None] = queue.Queue()
+        self._thread: threading.Thread | None = None
+        self._current_process: subprocess.Popen | None = None
+        self._closed = False
+        self._play_lock = threading.Lock()
+
+        self._platform = sys.platform
+        self._direct_playback = self._platform.startswith("win")
+
+        if self.enabled:
+            self._start()
+
+    def announce_count(self, count: int) -> None:
+        """将次数放入队列，后台线程播放对应音频"""
+        if not self.enabled or self._closed:
+            return
+        if count <= 0 or count > self.max_count:
+            return
+
+        requested_at = time.perf_counter()
+        if self._direct_playback:
+            audio_file = self._audio_path(count)
+            if not audio_file.exists():
+                logger.warning("Rep audio file missing: {}", audio_file)
+                return
+            try:
+                self._play(audio_file)
+                logger.info(
+                    "Rep audio submitted immediately: count={}, submit_ms={:.1f}",
+                    count,
+                    (time.perf_counter() - requested_at) * 1000,
+                )
+            except Exception as exc:
+                logger.warning("Failed to play rep count {}: {}", count, exc)
+            return
+
+        self._clear_pending_counts()
+        self._queue.put((count, requested_at))
+        logger.info("Rep audio queued: count={}", count)
+
+    def close(self) -> None:
+        """停止播报线程并释放资源"""
+        if not self.enabled or self._closed:
+            return
+
+        self._closed = True
+        self._queue.put(None)
+
+        if self._thread is not None:
+            self._thread.join(timeout=1.0)
+
+        self._stop_current_playback()
+        logger.info("Rep announcer closed")
+
+    def _start(self) -> None:
+        """启动后台播报线程"""
+        self.audio_dir.mkdir(parents=True, exist_ok=True)
+
+        if self._direct_playback:
+            import winsound
+
+            logger.info("Rep announcer initialized in direct Windows mode, audio_dir={}", self.audio_dir)
+            return
+
+        self._thread = threading.Thread(target=self._run, name="RepAnnouncer", daemon=True)
+        self._thread.start()
+
+        logger.info("Rep announcer initialized in queued mode, audio_dir={}", self.audio_dir)
+
+    def _run(self) -> None:
+        """后台线程：从队列取次数，播放对应音频文件"""
+        while True:
+            item = self._queue.get()
+            if item is None:
+                return
+            count, requested_at = item
+
+            audio_file = self._audio_path(count)
+            if not audio_file.exists():
+                logger.warning("Rep audio file missing: {}", audio_file)
+                continue
+
+            try:
+                self._play(audio_file)
+                logger.info(
+                    "Rep audio submitted from queue: count={}, queue_ms={:.1f}",
+                    count,
+                    (time.perf_counter() - requested_at) * 1000,
+                )
+            except Exception as exc:
+                logger.warning("Failed to play rep count {}: {}", count, exc)
+
+    def _play(self, audio_file: Path) -> None:
+        """播放音频文件（平台自适应）"""
+        with self._play_lock:
+            self._stop_current_playback()
+
+            if self._platform == "darwin":
+                self._current_process = subprocess.Popen(
+                    ["afplay", str(audio_file)],
+                    stdout=subprocess.DEVNULL,
+                    stderr=subprocess.DEVNULL,
+                )
+            elif self._platform.startswith("win"):
+                import winsound
+
+                winsound.PlaySound(
+                    str(audio_file),
+                    winsound.SND_FILENAME | winsound.SND_ASYNC | winsound.SND_NODEFAULT,
+                )
+            else:
+                player = shutil.which("paplay") or shutil.which("aplay")
+                if player is None:
+                    logger.warning("No audio player found")
+                    return
+                self._current_process = subprocess.Popen(
+                    [player, str(audio_file)],
+                    stdout=subprocess.DEVNULL,
+                    stderr=subprocess.DEVNULL,
+                )
+
+    def _stop_current_playback(self) -> None:
+        """中断当前正在播放的声音"""
+        if self._platform.startswith("win"):
+            try:
+                import winsound
+
+                winsound.PlaySound(None, winsound.SND_PURGE)
+            except Exception:
+                pass
+            return
+
+        if self._current_process is not None and self._current_process.poll() is None:
+            self._current_process.terminate()
+        self._current_process = None
+
+    def _audio_path(self, count: int) -> Path:
+        """获取某个次数对应的音频文件路径"""
+        suffix = ".aiff" if self._platform == "darwin" else ".wav"
+        return self.audio_dir / f"{count}{suffix}"
+
+    def _clear_pending_counts(self) -> None:
+        """清空队列中等待播放的次数，避免语音堆积"""
+        while True:
+            try:
+                item = self._queue.get_nowait()
+                if item is None:
+                    self._queue.put(None)
+                    return
+            except queue.Empty:
+                return

app/core/lifecycle.py

@@ -0,0 +1,22 @@
+from __future__ import annotations
+
+from app.diagnostics.crash_handler import enable_crash_handler
+from configs.load import config
+from app.audio.generate import generate_rep_audio_files
+
+def startup() -> None:
+    """应用启动初始化：开启崩溃日志和日志系统"""
+    enable_crash_handler(config.logging.dir_path)
+    from app.core.logging import setup_logging
+    setup_logging()
+
+    # 生成运动次数语音文件
+    generate_rep_audio_files(
+        max_count=config.audio.rep_max_count,
+        rate=config.audio.rep_announcer_rate,
+        output_dir=config.audio.resolved_audio_dir,
+        overwrite=False,
+        trim_leading_silence=config.audio.trim_leading_silence,
+        trim_silence_threshold=config.audio.trim_silence_threshold,
+        trim_silence_padding_ms=config.audio.trim_silence_padding_ms,
+    )

app/core/logging.py

@@ -0,0 +1,22 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+from loguru import logger
+
+from configs.load import config
+
+
+def setup_logging() -> None:
+    """配置loguru日志输出到按日期轮转的日志文件"""
+    log_dir = config.logging.dir_path
+    log_dir.mkdir(parents=True, exist_ok=True)
+
+    logger.add(
+        log_dir /"{time:YYYY-MM-DD}.log",
+        rotation=config.logging.rotation,
+        retention=config.logging.retention,
+        enqueue=True,
+        backtrace=True,
+        diagnose=True,
+    )

app/diagnostics/crash_handler.py

@@ -0,0 +1,12 @@
+from __future__ import annotations
+
+import faulthandler
+from pathlib import Path
+
+
+def enable_crash_handler(log_dir: str | Path) -> None:
+    """启用faulthandler，将崩溃堆栈写入日志文件"""
+    log_dir = Path(log_dir)
+    log_dir.mkdir(parents=True, exist_ok=True)
+    crash_log = open(log_dir / "posefit-crash.log", "a", buffering=1)
+    faulthandler.enable(file=crash_log, all_threads=True)

app/diagnostics/perf_timer.py

@@ -0,0 +1,39 @@
+from __future__ import annotations
+
+import time
+from contextlib import contextmanager
+
+from loguru import logger
+
+class PerfTimer:
+    """性能计时器，用于测量代码段执行耗时"""
+
+    def __init__(self, name: str = "") -> None:
+        self.name = name
+        self._start = 0.0
+        self._elapsed = 0.0
+
+    def start(self) -> PerfTimer:
+        """启动计时器"""
+        self._start = time.perf_counter()
+        return self
+
+    def stop(self) -> float:
+        """停止计时器并返回耗时（秒）"""
+        self._elapsed = time.perf_counter() - self._start
+        return self._elapsed
+
+    @property
+    def elapsed_ms(self) -> float:
+        """返回已记录耗时（毫秒）"""
+        return self._elapsed * 1000
+
+
+@contextmanager
+def measure(name: str = ""):
+    """上下文管理器：进入时计时，退出时记录耗时日志"""
+    timer = PerfTimer(name).start()
+    yield timer
+    elapsed = timer.stop()
+    if name:
+        logger.debug("{} took {:.1f}ms", name, timer.elapsed_ms)

app/exercises/dead_bug/detector.py

@@ -0,0 +1,254 @@
+from __future__ import annotations
+
+import threading
+import time
+
+import cv2
+import mediapipe as mp
+import numpy as np
+from loguru import logger
+
+from app.exercises.dead_bug.metrics import calculate_metrics
+from app.exercises.dead_bug.rules import has_required_visibility
+from app.exercises.dead_bug.state_machine import DeadBugStateMachine
+from app.exercises.dead_bug.types import DeadBugMetrics, DeadBugPhase, DeadBugResult, Point
+from app.rendering.overlay_renderer import draw_status_overlay
+from app.rendering.skeleton_renderer import draw_landmarks
+from app.vision.frame_utils import bgr_to_rgba
+from app.vision.pose_landmarker import PoseLandmarkerWrapper
+from app.vision.pose_types import (
+    LEFT_ANKLE,
+    LEFT_ELBOW,
+    LEFT_HIP,
+    LEFT_KNEE,
+    LEFT_SHOULDER,
+    LEFT_WRIST,
+    REQUIRED_LANDMARKS,
+    RIGHT_ANKLE,
+    RIGHT_ELBOW,
+    RIGHT_HIP,
+    RIGHT_KNEE,
+    RIGHT_SHOULDER,
+    RIGHT_WRIST,
+)
+
+class DeadBugDetector:
+    """死虫式（Dead Bug）运动检测器"""
+
+    def __init__(
+        self,
+        *,
+        model_path: str | None = None,
+        visibility_threshold: float = 0.45,
+        extension_confirm_frames: int = 4,
+        reset_confirm_frames: int = 3,
+        prefer_gpu: bool = True,
+    ) -> None:
+        """初始化姿态检测器、状态机和可视化渲染组件"""
+        self.visibility_threshold = visibility_threshold
+
+        self._latest_result = None
+        self._result_lock = threading.Lock()
+        self._result_event = threading.Event()
+        self._inflight = False
+        self._inflight_started_at = 0.0
+        self.last_timing: dict[str, float | bool] = {}
+
+        def on_result(pose_result, _image, _timestamp_ms):
+            with self._result_lock:
+                self._latest_result = pose_result
+                self._inflight = False
+                self._inflight_started_at = 0.0
+            self._result_event.set()
+
+        self._landmarker = PoseLandmarkerWrapper(
+            model_path=model_path,
+            prefer_gpu=prefer_gpu,
+            result_callback=on_result,
+        )
+
+        self._state = DeadBugStateMachine(
+            extension_confirm_frames=extension_confirm_frames,
+            reset_confirm_frames=reset_confirm_frames,
+        )
+
+        self._last_timestamp_ms = -1
+
+    def close(self) -> None:
+        """释放MediaPipe模型资源"""
+        self._landmarker.close()
+
+    def process_frame(self, bgr_frame: np.ndarray, timestamp_ms: int) -> tuple[np.ndarray, DeadBugResult]:
+        """处理单帧：姿态检测、指标计算、状态机更新、可视化叠加"""
+        total_started = time.perf_counter()
+        timestamp_ms = self._normalize_timestamp(timestamp_ms)
+        normalize_done = time.perf_counter()
+
+        with self._result_lock:
+            if self._inflight and time.monotonic() - self._inflight_started_at > 0.5:
+                logger.warning("MediaPipe detect_async timed out; allowing next frame submission")
+                self._inflight = False
+                self._inflight_started_at = 0.0
+            should_submit = not self._inflight
+            if should_submit:
+                self._inflight = True
+                self._inflight_started_at = time.monotonic()
+        lock_done = time.perf_counter()
+
+        if should_submit:
+            rgba_frame = bgr_to_rgba(bgr_frame)
+            convert_done = time.perf_counter()
+            mp_image = mp.Image(image_format=mp.ImageFormat.SRGBA, data=rgba_frame)
+            self._result_event.clear()
+            try:
+                self._landmarker.detect_async(mp_image, timestamp_ms)
+            except Exception:
+                with self._result_lock:
+                    self._inflight = False
+                    self._inflight_started_at = 0.0
+                raise
+            submit_done = time.perf_counter()
+            self._result_event.wait(timeout=0.08)
+            wait_done = time.perf_counter()
+        else:
+            convert_done = lock_done
+            submit_done = lock_done
+            wait_done = lock_done
+
+        with self._result_lock:
+            pose_result = self._latest_result
+        result_read_done = time.perf_counter()
+
+        annotated = bgr_frame.copy()
+        copy_done = time.perf_counter()
+
+        if pose_result is None or not pose_result.pose_landmarks:
+            self._state.mark_no_pose()
+            result = DeadBugResult(
+                rep_count=self._state.rep_count,
+                phase=DeadBugPhase.NO_POSE,
+                side=self._state.active_side,
+                is_standard=False,
+                feedback=["No full body detected"],
+                metrics=None,
+            )
+            draw_status_overlay(annotated, result)
+            self._record_timing(
+                total_started,
+                normalize_done,
+                lock_done,
+                convert_done,
+                submit_done,
+                wait_done,
+                result_read_done,
+                copy_done,
+                time.perf_counter(),
+                should_submit,
+            )
+            return annotated, result
+
+        landmarks = [Point(lm.x, lm.y, lm.z, getattr(lm, "visibility", 1.0)) for lm in pose_result.pose_landmarks[0]]
+        draw_landmarks(annotated, landmarks, REQUIRED_LANDMARKS, visibility_threshold=self.visibility_threshold)
+
+        if not has_required_visibility(landmarks, REQUIRED_LANDMARKS, self.visibility_threshold):
+            self._state.mark_no_pose()
+            result = DeadBugResult(
+                rep_count=self._state.rep_count,
+                phase=DeadBugPhase.NO_POSE,
+                side=self._state.active_side,
+                is_standard=False,
+                feedback=["Keep shoulders, elbows, wrists, hips, knees, ankles visible"],
+                metrics=None,
+            )
+            draw_status_overlay(annotated, result)
+            self._record_timing(
+                total_started,
+                normalize_done,
+                lock_done,
+                convert_done,
+                submit_done,
+                wait_done,
+                result_read_done,
+                copy_done,
+                time.perf_counter(),
+                should_submit,
+            )
+            return annotated, result
+
+        raw = calculate_metrics(
+            landmarks,
+            left_shoulder=LEFT_SHOULDER,
+            right_shoulder=RIGHT_SHOULDER,
+            left_elbow=LEFT_ELBOW,
+            right_elbow=RIGHT_ELBOW,
+            left_wrist=LEFT_WRIST,
+            right_wrist=RIGHT_WRIST,
+            left_hip=LEFT_HIP,
+            right_hip=RIGHT_HIP,
+            left_knee=LEFT_KNEE,
+            right_knee=RIGHT_KNEE,
+            left_ankle=LEFT_ANKLE,
+            right_ankle=RIGHT_ANKLE,
+            visibility_threshold=self.visibility_threshold,
+        )
+
+        metrics = DeadBugMetrics(
+            left_arm_extended=raw["left_arm_extended"],
+            right_arm_extended=raw["right_arm_extended"],
+            left_leg_extended=raw["left_leg_extended"],
+            right_leg_extended=raw["right_leg_extended"],
+            left_elbow_angle=raw["left_elbow_angle"],
+            right_elbow_angle=raw["right_elbow_angle"],
+            left_knee_angle=raw["left_knee_angle"],
+            right_knee_angle=raw["right_knee_angle"],
+            feedback=raw["feedback"],
+        )
+
+        result = self._state.update(metrics)
+        draw_status_overlay(annotated, result)
+        self._record_timing(
+            total_started,
+            normalize_done,
+            lock_done,
+            convert_done,
+            submit_done,
+            wait_done,
+            result_read_done,
+            copy_done,
+            time.perf_counter(),
+            should_submit,
+        )
+        return annotated, result
+
+    def _record_timing(
+        self,
+        total_started: float,
+        normalize_done: float,
+        lock_done: float,
+        convert_done: float,
+        submit_done: float,
+        wait_done: float,
+        result_read_done: float,
+        copy_done: float,
+        finished: float,
+        submitted: bool,
+    ) -> None:
+        self.last_timing = {
+            "total_ms": (finished - total_started) * 1000,
+            "timestamp_ms": (normalize_done - total_started) * 1000,
+            "lock_ms": (lock_done - normalize_done) * 1000,
+            "convert_ms": (convert_done - lock_done) * 1000,
+            "submit_ms": (submit_done - convert_done) * 1000,
+            "wait_ms": (wait_done - submit_done) * 1000,
+            "result_read_ms": (result_read_done - wait_done) * 1000,
+            "copy_ms": (copy_done - result_read_done) * 1000,
+            "postprocess_draw_ms": (finished - copy_done) * 1000,
+            "submitted": submitted,
+        }
+
+    def _normalize_timestamp(self, timestamp_ms: int) -> int:
+        """确保时间戳严格递增（MediaPipe要求）"""
+        if timestamp_ms <= self._last_timestamp_ms:
+            timestamp_ms = self._last_timestamp_ms + 1
+        self._last_timestamp_ms = timestamp_ms
+        return timestamp_ms

app/exercises/dead_bug/metrics.py

@@ -0,0 +1,95 @@
+from __future__ import annotations
+
+import cv2
+import numpy as np
+
+from app.exercises.dead_bug.types import Point
+
+
+def angle(a: Point, b: Point, c: Point) -> float:
+    """计算以b为顶点的三点夹角（度数）"""
+    ba = np.array([a.x - b.x, a.y - b.y], dtype=np.float32)
+    bc = np.array([c.x - b.x, c.y - b.y], dtype=np.float32)
+    denom = float(np.linalg.norm(ba) * np.linalg.norm(bc))
+    if denom == 0:
+        return 0.0
+    cos_value = float(np.dot(ba, bc) / denom)
+    return float(np.degrees(np.arccos(np.clip(cos_value, -1.0, 1.0))))
+
+
+def distance(a: Point, b: Point) -> float:
+    """计算两点之间的欧几里得距离（归一化坐标空间）"""
+    return float(np.hypot(a.x - b.x, a.y - b.y))
+
+
+def calculate_metrics(
+    lm: list[Point],
+    *,
+    left_shoulder: int,
+    right_shoulder: int,
+    left_elbow: int,
+    right_elbow: int,
+    left_wrist: int,
+    right_wrist: int,
+    left_hip: int,
+    right_hip: int,
+    left_knee: int,
+    right_knee: int,
+    left_ankle: int,
+    right_ankle: int,
+    visibility_threshold: float = 0.45,
+) -> dict:
+    """计算四肢关节角度、伸展状态及反馈信息"""
+    left_elbow_angle = angle(lm[left_shoulder], lm[left_elbow], lm[left_wrist])
+    right_elbow_angle = angle(lm[right_shoulder], lm[right_elbow], lm[right_wrist])
+    left_knee_angle = angle(lm[left_hip], lm[left_knee], lm[left_ankle])
+    right_knee_angle = angle(lm[right_hip], lm[right_knee], lm[right_ankle])
+
+    shoulder_width = distance(lm[left_shoulder], lm[right_shoulder])
+    hip_width = distance(lm[left_hip], lm[right_hip])
+    scale = max(shoulder_width, hip_width, 0.08)
+
+    left_arm_extended = (
+        left_elbow_angle >= 145
+        and distance(lm[left_shoulder], lm[left_wrist]) >= scale * 1.15
+        and lm[left_wrist].y <= lm[left_shoulder].y + scale * 0.35
+    )
+    right_arm_extended = (
+        right_elbow_angle >= 145
+        and distance(lm[right_shoulder], lm[right_wrist]) >= scale * 1.15
+        and lm[right_wrist].y <= lm[right_shoulder].y + scale * 0.35
+    )
+
+    left_leg_extended = (
+        left_knee_angle >= 150
+        and distance(lm[left_hip], lm[left_ankle]) >= scale * 1.55
+        and lm[left_ankle].y >= lm[left_knee].y - scale * 0.2
+    )
+    right_leg_extended = (
+        right_knee_angle >= 150
+        and distance(lm[right_hip], lm[right_ankle]) >= scale * 1.55
+        and lm[right_ankle].y >= lm[right_knee].y - scale * 0.2
+    )
+
+    feedback: list[str] = []
+    if left_arm_extended and left_elbow_angle < 160:
+        feedback.append("Straighten left arm")
+    if right_arm_extended and right_elbow_angle < 160:
+        feedback.append("Straighten right arm")
+    if left_leg_extended and left_knee_angle < 165:
+        feedback.append("Straighten left leg")
+    if right_leg_extended and right_knee_angle < 165:
+        feedback.append("Straighten right leg")
+
+    return {
+        "left_arm_extended": left_arm_extended,
+        "right_arm_extended": right_arm_extended,
+        "left_leg_extended": left_leg_extended,
+        "right_leg_extended": right_leg_extended,
+        "left_elbow_angle": left_elbow_angle,
+        "right_elbow_angle": right_elbow_angle,
+        "left_knee_angle": left_knee_angle,
+        "right_knee_angle": right_knee_angle,
+        "scale": scale,
+        "feedback": feedback,
+    }

app/exercises/dead_bug/rules.py

@@ -0,0 +1,27 @@
+from __future__ import annotations
+
+from app.exercises.dead_bug.types import DeadBugMetrics, Point
+
+
+def has_required_visibility(landmarks: list[Point], required_indices: tuple[int, ...], visibility_threshold: float) -> bool:
+    """检查所有必需关键点的可见度是否高于阈值"""
+    return all(landmarks[i].visibility >= visibility_threshold for i in required_indices)
+
+
+def detect_diagonal_extension(metrics: DeadBugMetrics) -> str | None:
+    """检测对角伸展（腿部只允许单侧伸展，手臂允许准备位上举带来的识别重叠）"""
+    if metrics.left_leg_extended and metrics.right_leg_extended:
+        return None
+
+    if metrics.right_leg_extended and metrics.left_arm_extended:
+        return "left_arm_right_leg"
+    if metrics.left_leg_extended and metrics.right_arm_extended:
+        return "right_arm_left_leg"
+    return None
+
+
+def is_ready_position(metrics: DeadBugMetrics) -> bool:
+    """判断是否处于准备姿态（双膝弯曲且双腿未伸展；dead bug 准备位允许手臂上举）"""
+    knees_bent = metrics.left_knee_angle <= 140 and metrics.right_knee_angle <= 140
+    legs_not_extended = not metrics.left_leg_extended and not metrics.right_leg_extended
+    return knees_bent and legs_not_extended

app/exercises/dead_bug/state_machine.py

@@ -0,0 +1,184 @@
+from __future__ import annotations
+
+from app.exercises.dead_bug.rules import detect_diagonal_extension, is_ready_position
+from app.exercises.dead_bug.types import DeadBugMetrics, DeadBugPhase, DeadBugResult
+
+_SMOOTHING_ALPHA = 0.45
+_TREND_DELTA_DEGREES = 1.5
+_SIDE_ANGLE_MARGIN = 10.0
+_EXTENSION_START_ANGLE = 125.0
+_EXTENSION_PEAK_ANGLE = 150.0
+_READY_KNEE_ANGLE = 140.0
+
+class DeadBugStateMachine:
+    """死虫式动作状态机：管理READY/EXTENDING/NEED_RESET/NO_POSE状态转换"""
+
+    def __init__(self, *, extension_confirm_frames: int = 4, reset_confirm_frames: int = 3) -> None:
+        """初始化并设置状态转换确认帧数"""
+        self.extension_confirm_frames = extension_confirm_frames
+        self.reset_confirm_frames = reset_confirm_frames
+
+        self.rep_count = 0
+        self.phase = DeadBugPhase.READY
+        self.active_side: str | None = None
+        self._candidate_side: str | None = None
+        self._candidate_frames = 0
+        self._reset_frames = 0
+        self._smooth_left_knee_angle: float | None = None
+        self._smooth_right_knee_angle: float | None = None
+        self._left_knee_delta = 0.0
+        self._right_knee_delta = 0.0
+
+    def mark_no_pose(self) -> None:
+        """姿态丢失时清掉候选帧；已确认的半程动作保留，等待重新可见后完成回收。"""
+        if self.phase == DeadBugPhase.READY:
+            self.phase = DeadBugPhase.NO_POSE
+            self.active_side = None
+        self._candidate_side = None
+        self._candidate_frames = 0
+        self._reset_frames = 0
+
+    def update(self, metrics: DeadBugMetrics) -> DeadBugResult:
+        """根据传入指标更新状态机并返回本次结果"""
+        self._update_knee_trends(metrics)
+
+        side = self._detect_motion_side(metrics)
+        ready = self._is_stable_ready(metrics)
+
+        if side is None:
+            self._candidate_side = None
+            self._candidate_frames = 0
+        elif side == self._candidate_side:
+            self._candidate_frames += 1
+        else:
+            self._candidate_side = side
+            self._candidate_frames = 1
+
+        if self.phase in (DeadBugPhase.READY, DeadBugPhase.NO_POSE):
+            if ready:
+                self.phase = DeadBugPhase.READY
+            if self._candidate_frames >= self.extension_confirm_frames and side is not None:
+                self.phase = DeadBugPhase.EXTENDING
+                self.active_side = side
+                self._reset_frames = 0
+        elif self.phase == DeadBugPhase.EXTENDING:
+            if ready or self._active_knee_retracting():
+                self.phase = DeadBugPhase.NEED_RESET
+                self._reset_frames = 1 if ready else 0
+        elif self.phase == DeadBugPhase.NEED_RESET:
+            if ready:
+                self._reset_frames += 1
+                if self._reset_frames >= self.reset_confirm_frames:
+                    self.rep_count += 1
+                    self.phase = DeadBugPhase.READY
+                    self.active_side = None
+                    self._candidate_side = None
+                    self._candidate_frames = 0
+                    self._reset_frames = 0
+            else:
+                self._reset_frames = 0
+
+        feedback = list(metrics.feedback)
+        display_side = detect_diagonal_extension(metrics)
+        if display_side is None and not ready:
+            feedback.append("Extend opposite arm and leg only")
+        if ready:
+            feedback.append("Ready position")
+        elif display_side == "left_arm_right_leg":
+            feedback.append("Left arm + right leg")
+        elif display_side == "right_arm_left_leg":
+            feedback.append("Right arm + left leg")
+
+        is_standard = display_side is not None and not metrics.feedback
+        return DeadBugResult(
+            rep_count=self.rep_count,
+            phase=self.phase,
+            side=display_side,
+            is_standard=is_standard,
+            feedback=feedback[:3],
+            metrics=metrics,
+        )
+
+    def _update_knee_trends(self, metrics: DeadBugMetrics) -> None:
+        """更新平滑膝角和本帧变化量，用连续趋势抵消单帧抖动。"""
+        previous_left = self._smooth_left_knee_angle
+        previous_right = self._smooth_right_knee_angle
+
+        if previous_left is None:
+            self._smooth_left_knee_angle = metrics.left_knee_angle
+            self._left_knee_delta = 0.0
+        else:
+            self._smooth_left_knee_angle = (
+                _SMOOTHING_ALPHA * metrics.left_knee_angle
+                + (1.0 - _SMOOTHING_ALPHA) * previous_left
+            )
+            self._left_knee_delta = self._smooth_left_knee_angle - previous_left
+
+        if previous_right is None:
+            self._smooth_right_knee_angle = metrics.right_knee_angle
+            self._right_knee_delta = 0.0
+        else:
+            self._smooth_right_knee_angle = (
+                _SMOOTHING_ALPHA * metrics.right_knee_angle
+                + (1.0 - _SMOOTHING_ALPHA) * previous_right
+            )
+            self._right_knee_delta = self._smooth_right_knee_angle - previous_right
+
+    def _detect_motion_side(self, metrics: DeadBugMetrics) -> str | None:
+        """基于膝角趋势推断正在伸展的对角侧，手臂只作为辅助校验。"""
+        raw_side = detect_diagonal_extension(metrics)
+        if raw_side is not None and self._side_has_extension_motion(raw_side):
+            return raw_side
+
+        left = self._smooth_left_knee_angle
+        right = self._smooth_right_knee_angle
+        if left is None or right is None:
+            return raw_side
+
+        both_legs_high = left >= _EXTENSION_PEAK_ANGLE and right >= _EXTENSION_PEAK_ANGLE
+        if both_legs_high and abs(left - right) < _SIDE_ANGLE_MARGIN:
+            return None
+
+        if (
+            right >= _EXTENSION_START_ANGLE
+            and right - left >= _SIDE_ANGLE_MARGIN
+            and (self._right_knee_delta >= _TREND_DELTA_DEGREES or right >= _EXTENSION_PEAK_ANGLE)
+        ):
+            return "left_arm_right_leg"
+        if (
+            left >= _EXTENSION_START_ANGLE
+            and left - right >= _SIDE_ANGLE_MARGIN
+            and (self._left_knee_delta >= _TREND_DELTA_DEGREES or left >= _EXTENSION_PEAK_ANGLE)
+        ):
+            return "right_arm_left_leg"
+        return raw_side
+
+    def _side_has_extension_motion(self, side: str) -> bool:
+        """确认对应腿处于伸展区或仍在伸展趋势中。"""
+        if side == "left_arm_right_leg":
+            angle = self._smooth_right_knee_angle
+            delta = self._right_knee_delta
+        else:
+            angle = self._smooth_left_knee_angle
+            delta = self._left_knee_delta
+        if angle is None:
+            return True
+        return angle >= _EXTENSION_PEAK_ANGLE or (
+            angle >= _EXTENSION_START_ANGLE and delta >= _TREND_DELTA_DEGREES
+        )
+
+    def _is_stable_ready(self, metrics: DeadBugMetrics) -> bool:
+        """准备位需要双腿回到屈膝区域；使用平滑膝角避免单帧阈值跳变。"""
+        left = self._smooth_left_knee_angle
+        right = self._smooth_right_knee_angle
+        if left is None or right is None:
+            return is_ready_position(metrics)
+        return left <= _READY_KNEE_ANGLE and right <= _READY_KNEE_ANGLE
+
+    def _active_knee_retracting(self) -> bool:
+        """确认已伸展的那条腿开始回收。"""
+        if self.active_side == "left_arm_right_leg":
+            return self._right_knee_delta <= -_TREND_DELTA_DEGREES
+        if self.active_side == "right_arm_left_leg":
+            return self._left_knee_delta <= -_TREND_DELTA_DEGREES
+        return False

app/exercises/dead_bug/types.py

@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from enum import Enum
+
+
+class DeadBugPhase(str, Enum):
+    """死虫式动作阶段枚举"""
+    READY = "ready"
+    EXTENDING = "extending"
+    NEED_RESET = "need_reset"
+    NO_POSE = "no_pose"
+
+
+@dataclass(frozen=True)
+class Point:
+    """三维关键点坐标及可见度"""
+    x: float
+    y: float
+    z: float
+    visibility: float
+
+
+@dataclass
+class DeadBugMetrics:
+    """四肢关节度量数据"""
+    left_arm_extended: bool
+    right_arm_extended: bool
+    left_leg_extended: bool
+    right_leg_extended: bool
+    left_elbow_angle: float
+    right_elbow_angle: float
+    left_knee_angle: float
+    right_knee_angle: float
+    feedback: list[str]
+
+
+@dataclass
+class DeadBugResult:
+    """单帧检测结果"""
+    rep_count: int
+    phase: DeadBugPhase
+    side: str | None
+    is_standard: bool
+    feedback: list[str]
+    metrics: DeadBugMetrics | None

app/main.py

@@ -0,0 +1,25 @@
+from __future__ import annotations
+
+import asyncio
+
+from loguru import logger
+
+from app.core.lifecycle import startup
+from app.signaling.websocket_server import main as serve
+
+
+def main():
+    """应用入口：启动服务并运行WebSocket信令服务器"""
+    startup()
+    logger.info("Starting server...")
+    try:
+        asyncio.run(serve())
+    except (KeyboardInterrupt, SystemExit):
+        logger.info("Server stopped by user")
+    except Exception as e:
+        logger.error(f"Server error: {e}")
+        raise
+
+
+if __name__ == "__main__":
+    main()

app/rendering/overlay_renderer.py

@@ -0,0 +1,21 @@
+from __future__ import annotations
+
+import cv2
+import numpy as np
+
+from app.exercises.dead_bug.types import DeadBugResult
+
+
+def draw_status_overlay(image: np.ndarray, result: DeadBugResult) -> None:
+    """在图像上叠加动作状态信息（次数、阶段、反馈）"""
+    color = (60, 220, 90) if result.is_standard else (50, 180, 255)
+    cv2.rectangle(image, (12, 12), (520, 142), (20, 20, 20), -1)
+    cv2.putText(image, f"Dead bug reps: {result.rep_count}", (28, 48), cv2.FONT_HERSHEY_SIMPLEX, 0.9, color, 2)
+    cv2.putText(image, f"phase: {result.phase.value}", (28, 82), cv2.FONT_HERSHEY_SIMPLEX, 0.68, (230, 230, 230), 2)
+    status = "standard" if result.is_standard else "adjust"
+    cv2.putText(image, f"status: {status}", (28, 116), cv2.FONT_HERSHEY_SIMPLEX, 0.68, color, 2)
+
+    y = 170
+    for text in result.feedback:
+        cv2.putText(image, text, (28, y), cv2.FONT_HERSHEY_SIMPLEX, 0.68, (255, 255, 255), 2)
+        y += 30

app/rendering/skeleton_renderer.py

@@ -0,0 +1,47 @@
+from __future__ import annotations
+
+import cv2
+import numpy as np
+
+from app.exercises.dead_bug.types import DeadBugResult, Point
+from app.vision.pose_types import _POSE_CONNECTIONS
+
+
+def draw_landmarks(
+    image: np.ndarray,
+    landmarks: list[Point],
+    required_indices: tuple[int, ...],
+    connections: tuple[tuple[int, int], ...] | None = None,
+    visibility_threshold: float = 0.45,
+    line_color: tuple[int, int, int] = (65, 180, 255),
+    point_color: tuple[int, int, int] = (80, 255, 120),
+    line_thickness: int = 2,
+    point_radius: int = 4,
+) -> None:
+    """绘制人体骨架关键点与连接线（仅绘制可见度达标的点）"""
+    if connections is None:
+        connections = _POSE_CONNECTIONS
+
+    h, w = image.shape[:2]
+
+    for start, end in connections:
+        if start >= len(landmarks) or end >= len(landmarks):
+            continue
+        p1 = landmarks[start]
+        p2 = landmarks[end]
+        if p1.visibility < visibility_threshold or p2.visibility < visibility_threshold:
+            continue
+        cv2.line(
+            image,
+            (int(p1.x * w), int(p1.y * h)),
+            (int(p2.x * w), int(p2.y * h)),
+            line_color,
+            line_thickness,
+        )
+
+    for idx in required_indices:
+        if idx >= len(landmarks):
+            continue
+        p = landmarks[idx]
+        if p.visibility >= visibility_threshold:
+            cv2.circle(image, (int(p.x * w), int(p.y * h)), point_radius, point_color, -1)

app/rendering/window_display.py

@@ -0,0 +1,25 @@
+from __future__ import annotations
+
+import cv2
+
+WINDOW_NAME = "Android Camera (WebRTC)"
+
+
+def show_frame(image, window_name: str = WINDOW_NAME) -> None:
+    """在OpenCV窗口中显示图像帧"""
+    cv2.imshow(window_name, image)
+
+
+def wait_key(delay_ms: int = 1) -> int:
+    """等待按键并返回ASCII码"""
+    return cv2.waitKey(delay_ms) & 0xFF
+
+
+def is_esc_pressed() -> bool:
+    """检测ESC键是否被按下"""
+    return wait_key(1) == 27
+
+
+def close_window() -> None:
+    """关闭所有OpenCV窗口"""
+    cv2.destroyAllWindows()

app/signaling/ice_parser.py

@@ -0,0 +1,31 @@
+from __future__ import annotations
+
+import re
+from typing import Any
+
+from aiortc import RTCIceCandidate
+
+
+def parse_ice(data: dict[str, Any]) -> RTCIceCandidate | None:
+    """解析ICE候选者字符串为RTCIceCandidate对象"""
+    match = re.match(
+        r'candidate:(\S+) (\d) (\S+) (\d+) (\S+) (\d+) typ (\S+)(?: raddr (\S+) rport (\d+))?',
+        data["candidate"],
+    )
+    if not match:
+        return None
+    g = match.groups()
+    cand = RTCIceCandidate(
+        foundation=g[0],
+        component=int(g[1]),
+        protocol=g[2].lower(),
+        priority=int(g[3]),
+        ip=g[4],
+        port=int(g[5]),
+        type=g[6],
+        relatedAddress=g[7],
+        relatedPort=int(g[8]) if g[8] else None,
+    )
+    cand.sdpMid = data.get("sdpMid")
+    cand.sdpMLineIndex = data.get("sdpMLineIndex", 0)
+    return cand

app/signaling/message_models.py

@@ -0,0 +1,13 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass
+class SignalingMessage:
+    """WebRTC信令消息数据模型"""
+    type: str
+    sdp: str = ""
+    candidate: str = ""
+    sdpMid: str | None = None
+    sdpMLineIndex: int = 0

app/signaling/websocket_server.py

@@ -0,0 +1,29 @@
+from __future__ import annotations
+
+import asyncio
+import json
+
+import websockets
+from loguru import logger
+
+from app.webrtc.peer_session import PeerSession
+from configs.load import config
+
+
+async def handle_client(websocket):
+    """处理单个WebSocket客户端连接"""
+    client = websocket.remote_address
+    logger.info(f"Client connected: {client}")
+
+    session = PeerSession()
+    await session.handle(websocket)
+
+    logger.info(f"Connection closed: {client}")
+
+
+async def main():
+    """启动WebSocket信令服务器"""
+    cfg = config.server
+    logger.info(f"WebRTC signaling server: ws://{cfg.host}:{cfg.port}")
+    async with websockets.serve(handle_client, cfg.host, cfg.port, max_size=cfg.max_ws_size):
+        await asyncio.Future()

app/vision/frame_utils.py

@@ -0,0 +1,26 @@
+from __future__ import annotations
+
+import cv2
+import numpy as np
+
+
+TARGET_WIDTH = 1280
+TARGET_HEIGHT = 720
+
+
+def resize_to_target(image: np.ndarray, width: int = TARGET_WIDTH, height: int = TARGET_HEIGHT) -> np.ndarray:
+    """将图像缩放到目标尺寸（仅当尺寸不一致时）"""
+    h, w = image.shape[:2]
+    if w == width and h == height:
+        return image
+    return cv2.resize(image, (width, height))
+
+
+def bgr_to_rgba(bgr: np.ndarray) -> np.ndarray:
+    """将BGR格式图像转换为RGBA格式"""
+    return cv2.cvtColor(bgr, cv2.COLOR_BGR2RGBA)
+
+
+def bgr_to_rgb(bgr: np.ndarray) -> np.ndarray:
+    """将BGR格式图像转换为RGB格式"""
+    return cv2.cvtColor(bgr, cv2.COLOR_BGR2RGB)

app/vision/pose_landmarker.py

@@ -0,0 +1,66 @@
+from __future__ import annotations
+
+import platform
+from typing import Callable
+
+import mediapipe as mp
+from loguru import logger
+
+from app.vision.pose_models import DEFAULT_MODEL_PATH
+
+PoseLandmarker = mp.tasks.vision.PoseLandmarker
+PoseLandmarkerOptions = mp.tasks.vision.PoseLandmarkerOptions
+VisionRunningMode = mp.tasks.vision.RunningMode
+BaseOptions = mp.tasks.BaseOptions
+
+class PoseLandmarkerWrapper:
+    """MediaPipe姿态关键点检测器封装"""
+
+    def __init__(
+        self,
+        *,
+        model_path: str | None = None,
+        prefer_gpu: bool = True,
+        result_callback: Callable | None = None,
+    ) -> None:
+        """初始化姿态检测器，优先尝试GPU委托，失败则回退到CPU"""
+        self.model_path = model_path or DEFAULT_MODEL_PATH
+
+        if prefer_gpu:
+            try:
+                if platform.system() == "Windows":
+                    logger.warning(
+                        "MediaPipe GPU delegate requested, but MediaPipe Tasks Python does not support GPU delegate on Windows; "
+                        "Intel Iris Xe cannot be used by this backend and CPU fallback is expected"
+                    )
+                self.delegate = BaseOptions.Delegate.GPU
+                self._landmarker = self._create(self.delegate, result_callback)
+                logger.info("MediaPipe PoseLandmarker initialized with GPU delegate")
+                return
+            except Exception as exc:
+                logger.warning("MediaPipe GPU delegate unavailable, falling back to CPU: {}", exc)
+
+        self.delegate = BaseOptions.Delegate.CPU
+        self._landmarker = self._create(self.delegate, result_callback)
+        logger.info("MediaPipe PoseLandmarker initialized with CPU delegate")
+
+    def _create(self, delegate, result_callback=None):
+        """根据委托类型和回调创建PoseLandmarker实例"""
+        options = PoseLandmarkerOptions(
+            base_options=BaseOptions(model_asset_path=self.model_path, delegate=delegate),
+            running_mode=VisionRunningMode.LIVE_STREAM,
+            result_callback=result_callback,
+            num_poses=1,
+            min_pose_detection_confidence=0.5,
+            min_pose_presence_confidence=0.5,
+            min_tracking_confidence=0.5,
+        )
+        return PoseLandmarker.create_from_options(options)
+
+    def detect_async(self, mp_image, timestamp_ms: int) -> None:
+        """异步执行姿态检测"""
+        return self._landmarker.detect_async(mp_image, timestamp_ms)
+
+    def close(self) -> None:
+        """释放MediaPipe资源"""
+        self._landmarker.close()

app/vision/pose_models.py

@@ -0,0 +1,14 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import mediapipe as mp
+
+BaseOptions = mp.tasks.BaseOptions
+
+_MODELS_DIR = Path(__file__).resolve().parent.parent.parent / "pose_models"
+
+POSE_LANDMARKER_FULL = _MODELS_DIR / "pose_landmarker_full.task"
+POSE_LANDMARKER_LITE = _MODELS_DIR / "pose_landmarker_lite.task"
+
+DEFAULT_MODEL_PATH = str(POSE_LANDMARKER_FULL) if POSE_LANDMARKER_FULL.exists() else str(POSE_LANDMARKER_LITE)

app/vision/pose_types.py

@@ -0,0 +1,80 @@
+from __future__ import annotations
+
+_POSE_CONNECTIONS = (
+    (11, 12),
+    (11, 13),
+    (13, 15),
+    (12, 14),
+    (14, 16),
+    (11, 23),
+    (12, 24),
+    (23, 24),
+    (23, 25),
+    (25, 27),
+    (24, 26),
+    (26, 28),
+)
+
+LANDMARK_NAMES: dict[int, str] = {
+    0: "nose",
+    1: "left_eye_inner",
+    2: "left_eye",
+    3: "left_eye_outer",
+    4: "right_eye_inner",
+    5: "right_eye",
+    6: "right_eye_outer",
+    7: "left_ear",
+    8: "right_ear",
+    9: "mouth_left",
+    10: "mouth_right",
+    11: "left_shoulder",
+    12: "right_shoulder",
+    13: "left_elbow",
+    14: "right_elbow",
+    15: "left_wrist",
+    16: "right_wrist",
+    17: "left_pinky",
+    18: "right_pinky",
+    19: "left_index",
+    20: "right_index",
+    21: "left_thumb",
+    22: "right_thumb",
+    23: "left_hip",
+    24: "right_hip",
+    25: "left_knee",
+    26: "right_knee",
+    27: "left_ankle",
+    28: "right_ankle",
+    29: "left_heel",
+    30: "right_heel",
+    31: "left_foot_index",
+    32: "right_foot_index",
+}
+
+LEFT_SHOULDER = 11
+RIGHT_SHOULDER = 12
+LEFT_ELBOW = 13
+RIGHT_ELBOW = 14
+LEFT_WRIST = 15
+RIGHT_WRIST = 16
+LEFT_HIP = 23
+RIGHT_HIP = 24
+LEFT_KNEE = 25
+RIGHT_KNEE = 26
+LEFT_ANKLE = 27
+RIGHT_ANKLE = 28
+
+REQUIRED_LANDMARKS = (
+    LEFT_SHOULDER,
+    RIGHT_SHOULDER,
+    LEFT_ELBOW,
+    RIGHT_ELBOW,
+    LEFT_WRIST,
+    RIGHT_WRIST,
+    LEFT_HIP,
+    RIGHT_HIP,
+    LEFT_KNEE,
+    RIGHT_KNEE,
+    LEFT_ANKLE,
+    RIGHT_ANKLE,
+)

app/webrtc/frame_source.py

@@ -0,0 +1,15 @@
+from __future__ import annotations
+
+import numpy as np
+from loguru import logger
+
+
+TARGET_WIDTH = 1280
+TARGET_HEIGHT = 720
+
+
+def validate_frame_size(image: np.ndarray, width: int = TARGET_WIDTH, height: int = TARGET_HEIGHT) -> None:
+    """验证视频帧尺寸是否与目标尺寸一致，不一致时记录警告"""
+    h, w = image.shape[:2]
+    if w != width or h != height:
+        logger.warning("Unexpected frame size: {}x{}", w, h)

app/webrtc/peer_session.py

@@ -0,0 +1,74 @@
+from __future__ import annotations
+
+import asyncio
+import json
+
+import websockets
+from aiortc import RTCPeerConnection, RTCSessionDescription
+from loguru import logger
+
+from app.signaling.ice_parser import parse_ice
+from app.webrtc.video_receiver import VideoReceiver
+
+class PeerSession:
+    """WebRTC对等连接会话管理"""
+
+    def __init__(self) -> None:
+        self._pc = RTCPeerConnection()
+        self._video_task: asyncio.Task | None = None
+
+    async def handle(self, websocket) -> None:
+        """处理WebSocket信令交互与WebRTC连接建立"""
+        self._setup_events()
+
+        try:
+            async for message in websocket:
+                data = json.loads(message)
+                msg_type = data.get("type")
+
+                if msg_type == "offer":
+                    offer = RTCSessionDescription(sdp=data["sdp"], type="offer")
+                    await self._pc.setRemoteDescription(offer)
+                    answer = await self._pc.createAnswer()
+                    await self._pc.setLocalDescription(answer)
+                    await websocket.send(json.dumps({
+                        "type": "answer",
+                        "sdp": self._pc.localDescription.sdp,
+                    }))
+
+                elif msg_type == "candidate":
+                    cand = parse_ice(data)
+                    if cand:
+                        await self._pc.addIceCandidate(cand)
+
+        except websockets.ConnectionClosed:
+            pass
+        except Exception as e:
+            logger.exception(f"Error: {e}")
+        finally:
+            await self._cleanup()
+
+    def _setup_events(self) -> None:
+        """注册ICE连接状态变化和视频轨道接收事件处理器"""
+        @self._pc.on("track")
+        async def on_track(track):
+            logger.info(f"Track received: kind={track.kind}")
+            if track.kind == "video":
+                receiver = VideoReceiver(track)
+                self._video_task = asyncio.ensure_future(receiver.run())
+
+        @self._pc.on("iceconnectionstatechange")
+        async def on_iceconnectionstatechange():
+            logger.info(f"ICE state: {self._pc.iceConnectionState}")
+            if self._pc.iceConnectionState in ("failed", "closed", "disconnected"):
+                await self._pc.close()
+
+    async def _cleanup(self) -> None:
+        """清理视频任务并关闭对等连接"""
+        if self._video_task:
+            self._video_task.cancel()
+            try:
+                await self._video_task
+            except asyncio.CancelledError:
+                pass
+        await self._pc.close()

app/webrtc/video_receiver.py

@@ -0,0 +1,213 @@
+from __future__ import annotations
+
+import asyncio
+import time
+
+import cv2
+from aiortc.mediastreams import MediaStreamError
+from loguru import logger
+
+from app.audio.rep_announcer import RepAnnouncer
+from app.exercises.dead_bug.detector import DeadBugDetector
+from app.rendering.window_display import close_window, is_esc_pressed, show_frame
+from configs.load import config
+
+
+def _format_pose_debug(pose_result) -> str:
+    """格式化姿态检测结果用于调试日志输出"""
+    metrics = pose_result.metrics
+    if metrics is None:
+        return "metrics=None"
+    return (
+        f"side={pose_result.side}, standard={pose_result.is_standard}, "
+        f"angles(le={metrics.left_elbow_angle:.1f}, re={metrics.right_elbow_angle:.1f}, "
+        f"lk={metrics.left_knee_angle:.1f}, rk={metrics.right_knee_angle:.1f}), "
+        f"extended(la={metrics.left_arm_extended}, ra={metrics.right_arm_extended}, "
+        f"ll={metrics.left_leg_extended}, rl={metrics.right_leg_extended})"
+    )
+
+
+def _new_perf_window() -> dict:
+    return {
+        "frames": 0,
+        "processed": 0,
+        "loop_ms": 0.0,
+        "to_ndarray_ms": 0.0,
+        "detect_ms": 0.0,
+        "show_ms": 0.0,
+        "max_loop_ms": 0.0,
+        "max_detect_ms": 0.0,
+        "detector": {},
+    }
+
+
+def _add_detector_timing(perf: dict, timing: dict[str, float | bool]) -> None:
+    detector = perf["detector"]
+    for key, value in timing.items():
+        if key == "submitted":
+            detector[key] = detector.get(key, 0) + (1 if value else 0)
+            continue
+        value = float(value)
+        detector[key] = detector.get(key, 0.0) + value
+        max_key = f"max_{key}"
+        detector[max_key] = max(detector.get(max_key, 0.0), value)
+
+
+def _avg(perf: dict, key: str, denominator: int) -> float:
+    if denominator <= 0:
+        return 0.0
+    return perf.get(key, 0.0) / denominator
+
+
+class VideoReceiver:
+    """视频轨道接收与运动检测流水线"""
+
+    def __init__(self, track) -> None:
+        self._track = track
+
+    async def run(self) -> None:
+        """持续接收视频帧并进行姿态检测、渲染和语音播报"""
+        log_every_n_frames = max(1, config.video.log_every_n_frames)
+        perf_log_every_n_frames = max(1, config.video.perf_log_every_n_frames)
+        slow_frame_ms = max(0.0, config.video.slow_frame_ms)
+        logger.info(
+            "Start receiving video frames, process_every_n={}, log_every_n={}, perf_log_every_n={}, slow_frame_ms={}",
+            config.video.process_every_n_frames,
+            log_every_n_frames,
+            perf_log_every_n_frames,
+            slow_frame_ms,
+        )
+        logger.info(
+            "OpenCV OpenCL status: have_opencl={}, use_opencl={}",
+            cv2.ocl.haveOpenCL(),
+            cv2.ocl.useOpenCL(),
+        )
+
+        frame_count = 0
+        processed_count = 0
+        detector = DeadBugDetector(
+            model_path=config.model.resolved_path,
+            visibility_threshold=config.dead_bug.visibility_threshold,
+            extension_confirm_frames=config.dead_bug.extension_confirm_frames,
+            reset_confirm_frames=config.dead_bug.reset_confirm_frames,
+            prefer_gpu=config.model.prefer_gpu,
+        )
+        announcer = RepAnnouncer(
+            enabled=config.audio.rep_announcer_enabled,
+            max_count=config.audio.rep_max_count,
+            audio_dir=config.audio.resolved_audio_dir,
+        )
+        last_announced_rep = 0
+        last_pose_result = None
+        last_annotated = None
+        perf = _new_perf_window()
+
+        try:
+            while True:
+                loop_started = time.perf_counter()
+                frame = await self._track.recv()
+                frame_count += 1
+                recv_done = time.perf_counter()
+                raw_img = frame.to_ndarray(format="bgr24")
+                ndarray_done = time.perf_counter()
+                timestamp_ms = int(frame.time * 1000) if frame.time is not None else frame_count * 33
+
+                detect_ms = 0.0
+                if frame_count % config.video.process_every_n_frames == 0 or last_pose_result is None:
+                    detect_started = time.perf_counter()
+                    processed_count += 1
+                    last_annotated, last_pose_result = detector.process_frame(raw_img, timestamp_ms)
+                    detect_ms = (time.perf_counter() - detect_started) * 1000
+                    perf["processed"] += 1
+                    perf["detect_ms"] += detect_ms
+                    perf["max_detect_ms"] = max(perf["max_detect_ms"], detect_ms)
+                    _add_detector_timing(perf, detector.last_timing)
+                    if last_pose_result.rep_count > last_announced_rep:
+                        last_announced_rep = last_pose_result.rep_count
+                        announce_started = time.perf_counter()
+                        announcer.announce_count(last_announced_rep)
+                        logger.info(
+                            "Rep completed and audio requested: count={}, frame={}, announce_call_ms={:.1f}",
+                            last_announced_rep,
+                            frame_count,
+                            (time.perf_counter() - announce_started) * 1000,
+                        )
+
+                display_img = last_annotated if last_annotated is not None else raw_img
+                show_started = time.perf_counter()
+                show_frame(display_img)
+                show_done = time.perf_counter()
+
+                if frame_count % log_every_n_frames == 0:
+                    logger.info(
+                        "Received {} frames, processed={}, raw_shape={}, reps={}, phase={}, feedback={}, {}",
+                        frame_count,
+                        processed_count,
+                        raw_img.shape,
+                        last_pose_result.rep_count if last_pose_result is not None else 0,
+                        last_pose_result.phase.value if last_pose_result is not None else "none",
+                        " | ".join(last_pose_result.feedback) if last_pose_result is not None else "",
+                        _format_pose_debug(last_pose_result) if last_pose_result is not None else "metrics=None",
+                    )
+
+                loop_ms = (show_done - loop_started) * 1000
+                to_ndarray_ms = (ndarray_done - recv_done) * 1000
+                show_ms = (show_done - show_started) * 1000
+                perf["frames"] += 1
+                perf["loop_ms"] += loop_ms
+                perf["to_ndarray_ms"] += to_ndarray_ms
+                perf["show_ms"] += show_ms
+                perf["max_loop_ms"] = max(perf["max_loop_ms"], loop_ms)
+
+                if slow_frame_ms and loop_ms >= slow_frame_ms:
+                    logger.warning(
+                        "Slow video frame: frame={}, loop_ms={:.1f}, detect_ms={:.1f}, to_ndarray_ms={:.1f}, show_ms={:.1f}, shape={}",
+                        frame_count,
+                        loop_ms,
+                        detect_ms,
+                        to_ndarray_ms,
+                        show_ms,
+                        raw_img.shape,
+                    )
+
+                if frame_count % perf_log_every_n_frames == 0:
+                    frames = perf["frames"]
+                    processed = perf["processed"]
+                    detector_perf = perf["detector"]
+                    logger.info(
+                        "Perf window: frames={}, processed={}, avg_loop_ms={:.1f}, max_loop_ms={:.1f}, avg_to_ndarray_ms={:.1f}, "
+                        "avg_detect_ms={:.1f}, max_detect_ms={:.1f}, avg_show_ms={:.1f}, detector_avg_total_ms={:.1f}, "
+                        "detector_max_total_ms={:.1f}, detector_avg_wait_ms={:.1f}, detector_max_wait_ms={:.1f}, "
+                        "detector_avg_convert_ms={:.1f}, detector_avg_postprocess_draw_ms={:.1f}, detector_submitted={}",
+                        frames,
+                        processed,
+                        _avg(perf, "loop_ms", frames),
+                        perf["max_loop_ms"],
+                        _avg(perf, "to_ndarray_ms", frames),
+                        _avg(perf, "detect_ms", processed),
+                        perf["max_detect_ms"],
+                        _avg(perf, "show_ms", frames),
+                        _avg(detector_perf, "total_ms", processed),
+                        detector_perf.get("max_total_ms", 0.0),
+                        _avg(detector_perf, "wait_ms", processed),
+                        detector_perf.get("max_wait_ms", 0.0),
+                        _avg(detector_perf, "convert_ms", processed),
+                        _avg(detector_perf, "postprocess_draw_ms", processed),
+                        detector_perf.get("submitted", 0),
+                    )
+                    perf = _new_perf_window()
+
+                if is_esc_pressed():
+                    logger.info("ESC pressed, closing display")
+                    break
+
+        except asyncio.CancelledError:
+            logger.info("Video receive task cancelled")
+        except MediaStreamError:
+            logger.info("Video track ended")
+        except Exception as e:
+            logger.exception(f"Video receive error: {e!r}")
+        finally:
+            announcer.close()
+            detector.close()
+            close_window()

config.yaml

@@ -0,0 +1,37 @@
+# PoseFit Server Configuration
+# ============================
+
+server:
+  host: "0.0.0.0"
+  port: 8765
+  max_ws_size: 10485760   # 10 MB
+
+video:
+  process_every_n_frames: 2
+  log_every_n_frames: 30
+  perf_log_every_n_frames: 30
+  slow_frame_ms: 100
+
+model:
+  path: "./pose_models/pose_landmarker_full.task"
+  prefer_gpu: true
+
+dead_bug:
+  visibility_threshold: 0.45
+  extension_confirm_frames: 4
+  reset_confirm_frames: 3
+
+audio:
+  rep_announcer_enabled: true
+  rep_announcer_rate: 185
+  rep_announcer_volume: 1.0
+  rep_max_count: 200              # 预生成语音文件的最大次数
+  rep_audio_dir: ""               # 空 = 自动使用 app/audio/reps
+  trim_leading_silence: true
+  trim_silence_threshold: 500
+  trim_silence_padding_ms: 20
+
+logging:
+  dir: logs
+  rotation: "20 MB"
+  retention: "14 days"

configs/load.py

@@ -0,0 +1,60 @@
+from __future__ import annotations
+
+import dataclasses
+from pathlib import Path
+from typing import Any
+
+import yaml
+
+from configs.models import (
+    AppConfig,
+    AudioConfig,
+    DeadBugConfig,
+    LoggingConfig,
+    ModelConfig,
+    ServerConfig,
+    VideoConfig,
+)
+
+_PROJECT_ROOT = Path(__file__).resolve().parent.parent
+
+_SECTION_CLASS = {
+    "server": ServerConfig,
+    "video": VideoConfig,
+    "model": ModelConfig,
+    "dead_bug": DeadBugConfig,
+    "audio": AudioConfig,
+    "logging": LoggingConfig,
+}
+
+
+def _dict_to_dataclass(cls: type, data: dict[str, Any] | None) -> dict[str, Any]:
+    """将字典过滤为仅包含指定dataclass字段的键值对"""
+    if data is None:
+        return {}
+    field_names = {f.name for f in dataclasses.fields(cls)}
+    return {k: v for k, v in data.items() if k in field_names}
+
+
+def _read_yaml(path: Path) -> dict[str, Any]:
+    """读取YAML配置文件并返回字典"""
+    if not path.exists():
+        return {}
+    with open(path, encoding="utf-8") as f:
+        return yaml.safe_load(f) or {}
+
+
+def load_config(config_path: str | Path | None = None) -> AppConfig:
+    """加载并解析应用配置，返回AppConfig实例"""
+    if config_path is None:
+        config_path = _PROJECT_ROOT / "config.yaml"
+
+    raw = _read_yaml(Path(config_path))
+
+    return AppConfig(**{
+        section: cls(**_dict_to_dataclass(cls, raw.get(section)))
+        for section, cls in _SECTION_CLASS.items()
+    })
+
+
+config = load_config()

configs/models.py

@@ -0,0 +1,87 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+
+
+@dataclass
+class ServerConfig:
+    """WebSocket服务器配置"""
+    host: str = "0.0.0.0"
+    port: int = 8765
+    max_ws_size: int = 10_485_760
+
+
+@dataclass
+class VideoConfig:
+    """视频帧处理配置"""
+    process_every_n_frames: int = 2
+    log_every_n_frames: int = 30
+    perf_log_every_n_frames: int = 30
+    slow_frame_ms: float = 100.0
+
+
+@dataclass
+class ModelConfig:
+    """姿态检测模型配置"""
+    path: str = ""
+    prefer_gpu: bool = True
+
+    @property
+    def resolved_path(self) -> str:
+        """返回模型文件的绝对路径"""
+        if self.path:
+            return self.path
+        return str(Path(__file__).resolve().parent.parent / "pose_models" / "pose_landmarker_full.task")
+
+
+@dataclass
+class DeadBugConfig:
+    """死虫式（Dead Bug）运动检测配置"""
+    visibility_threshold: float = 0.45
+    extension_confirm_frames: int = 4
+    reset_confirm_frames: int = 3
+
+
+@dataclass
+class AudioConfig:
+    """语音播报配置"""
+    rep_announcer_enabled: bool = True
+    rep_announcer_rate: int = 185
+    rep_announcer_volume: float = 1.0
+    rep_max_count: int = 200
+    rep_audio_dir: str = ""
+    trim_leading_silence: bool = True
+    trim_silence_threshold: int = 500
+    trim_silence_padding_ms: int = 20
+
+    @property
+    def resolved_audio_dir(self) -> Path:
+        """返回语音文件目录的绝对路径"""
+        if self.rep_audio_dir:
+            return Path(self.rep_audio_dir)
+        return Path(__file__).resolve().parent.parent / "resources" / "audio" / "reps"
+
+
+@dataclass
+class LoggingConfig:
+    """日志配置"""
+    dir: str = "logs"
+    rotation: str = "20 MB"
+    retention: str = "14 days"
+
+    @property
+    def dir_path(self) -> Path:
+        """返回日志目录的绝对路径"""
+        return Path(__file__).resolve().parent.parent / self.dir
+
+
+@dataclass
+class AppConfig:
+    """应用总配置，聚合所有子配置"""
+    server: ServerConfig = field(default_factory=ServerConfig)
+    video: VideoConfig = field(default_factory=VideoConfig)
+    model: ModelConfig = field(default_factory=ModelConfig)
+    dead_bug: DeadBugConfig = field(default_factory=DeadBugConfig)
+    audio: AudioConfig = field(default_factory=AudioConfig)
+    logging: LoggingConfig = field(default_factory=LoggingConfig)

handle_client.py

@@ -1,121 +0,0 @@
-import asyncio
-import json
-import re
-import websockets
-import cv2
-from loguru import logger
-from aiortc import RTCPeerConnection, RTCSessionDescription, RTCIceCandidate
-
-
-async def handle_client(websocket):
-    client = websocket.remote_address
-    logger.info(f"Client connected: {client}")
-
-    pc = RTCPeerConnection()
-    video_task = None
-
-    def parse_ice(data):
-        match = re.match(
-            r'candidate:(\S+) (\d) (\S+) (\d+) (\S+) (\d+) typ (\S+)(?: raddr (\S+) rport (\d+))?',
-            data["candidate"]
-        )
-        if not match:
-            return None
-        g = match.groups()
-        cand = RTCIceCandidate(
-            foundation=g[0],
-            component=int(g[1]),
-            protocol=g[2].lower(),
-            priority=int(g[3]),
-            ip=g[4],
-            port=int(g[5]),
-            type=g[6],
-            relatedAddress=g[7],
-            relatedPort=int(g[8]) if g[8] else None,
-        )
-        cand.sdpMid = data.get("sdpMid")
-        cand.sdpMLineIndex = data.get("sdpMLineIndex", 0)
-        return cand
-
-    async def receive_video(track):
-        logger.info("Start receiving video frames")
-        frame_count = 0
-        try:
-            while True:
-                frame = await track.recv()
-                frame_count += 1
-                img = frame.to_ndarray(format="bgr24")
-                cv2.imshow("Android Camera (WebRTC)", img)
-
-                if frame_count % 100 == 0:
-                    logger.info(f"Received {frame_count} frames, shape={img.shape}")
-
-                if cv2.waitKey(1) & 0xFF == 27:
-                    logger.info("ESC pressed, closing display")
-                    break
-        except asyncio.CancelledError:
-            logger.info("Video receive task cancelled")
-        except Exception as e:
-            logger.error(f"Video receive error: {e}")
-
-    @pc.on("track")
-    async def on_track(track):
-        logger.info(f"Track received: kind={track.kind}")
-        if track.kind == "video":
-            nonlocal video_task
-            video_task = asyncio.ensure_future(receive_video(track))
-
-    @pc.on("iceconnectionstatechange")
-    async def on_iceconnectionstatechange():
-        logger.info(f"ICE state: {pc.iceConnectionState}")
-        if pc.iceConnectionState in ("failed", "closed", "disconnected"):
-            await pc.close()
-
-    try:
-        async for message in websocket:
-            data = json.loads(message)
-            msg_type = data.get("type")
-
-            if msg_type == "offer":
-                offer = RTCSessionDescription(sdp=data["sdp"], type="offer")
-                await pc.setRemoteDescription(offer)
-
-                answer = await pc.createAnswer()
-                await pc.setLocalDescription(answer)
-
-                await websocket.send(json.dumps({
-                    "type": "answer",
-                    "sdp": pc.localDescription.sdp,
-                }))
-
-            elif msg_type == "candidate":
-                cand = parse_ice(data)
-                if cand:
-                    await pc.addIceCandidate(cand)
-
-    except websockets.ConnectionClosed:
-        logger.info(f"Client disconnected: {client}")
-    except Exception as e:
-        logger.exception(f"Error: {e}")
-    finally:
-        if video_task:
-            video_task.cancel()
-            try:
-                await video_task
-            except asyncio.CancelledError:
-                pass
-        await pc.close()
-        cv2.destroyAllWindows()
-        logger.info(f"Connection closed: {client}")
-
-
-async def main():
-    host = "0.0.0.0"
-    port = 8765
-    logger.info(f"WebRTC signaling server: ws://{host}:{port}")
-    async with websockets.serve(handle_client, host, port, max_size=10 * 1024 * 1024):
-        await asyncio.Future()
-
-
-if __name__ == "__main__":
-    asyncio.run(main())

main.py

@@ -1,27 +0,0 @@
-import mediapipe as mp
-
-def main():
-    BaseOptions = mp.tasks.BaseOptions
-    PoseLandmarker = mp.tasks.vision.PoseLandmarker
-    PoseLandmarkerOptions = mp.tasks.vision.PoseLandmarkerOptions
-    PoseLandmarkerResult = mp.tasks.vision.PoseLandmarkerResult
-    VisionRunningMode = mp.tasks.vision.RunningMode
-
-    # Create a pose landmarker instance with the live stream mode:
-    def print_result(result: PoseLandmarkerResult, output_image: mp.Image, timestamp_ms: int):
-        print('pose landmarker result: {}'.format(result))
-
-    options = PoseLandmarkerOptions(
-        base_options=BaseOptions(model_asset_path=model_path),
-        running_mode=VisionRunningMode.LIVE_STREAM,
-        result_callback=print_result)
-
-    with PoseLandmarker.create_from_options(options) as landmarker:
-        
-
-
-
-
-
-if __name__ == '__main__':
-    main()

requirements.txt

@@ -1,5 +1,8 @@
 aiortc>=1.9.0
 websockets>=13.0
-opencv-python>=4.10.0
-numpy>=2.0.0
+opencv-contrib-python>=4.10.0
+numpy>=1.26,<2
 loguru>=0.7.0
+mediapipe==0.10.21
+pyttsx3>=2.99
+pyyaml>=6.0

run.py

@@ -0,0 +1,4 @@
+from app.main import main
+
+if __name__ == "__main__":
+    main()

tests/test_dead_bug_rules.py

@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+from app.exercises.dead_bug.rules import detect_diagonal_extension, has_required_visibility, is_ready_position
+from app.exercises.dead_bug.types import DeadBugMetrics, Point
+
+class TestDeadBugRules:
+    """死虫式规则函数单元测试"""
+
+    def _make_landmark(self, x=0.5, y=0.5, z=0.0, visibility=1.0):
+        """创建测试用Point对象"""
+        return Point(x, y, z, visibility)
+
+    def _make_visible_landmarks(self):
+        """创建33个全可见的测试用关键点"""
+        return [self._make_landmark() for _ in range(33)]
+
+    def test_has_required_visibility_all_visible(self):
+        """测试：所有关键点可见时应返回True"""
+        lm = self._make_visible_landmarks()
+        indices = (11, 12, 13, 14, 15, 16, 23, 24, 25, 26, 27, 28)
+        assert has_required_visibility(lm, indices, 0.45)
+
+    def test_has_required_visibility_low(self):
+        """测试：关键点可见度过低时应返回False"""
+        lm = self._make_visible_landmarks()
+        lm[11] = self._make_landmark(visibility=0.1)
+        indices = (11, 12, 13, 14, 15, 16, 23, 24, 25, 26, 27, 28)
+        assert not has_required_visibility(lm, indices, 0.45)
+
+    def test_detect_diagonal_extension_none(self):
+        """测试：四肢均未伸展时应返回None"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=False, right_arm_extended=False,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=90, right_elbow_angle=90,
+            left_knee_angle=90, right_knee_angle=90,
+            feedback=[],
+        )
+        assert detect_diagonal_extension(metrics) is None
+
+    def test_detect_diagonal_extension_left_arm_right_leg(self):
+        """测试：左臂+右腿对角伸展检测"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=False,
+            left_leg_extended=False, right_leg_extended=True,
+            left_elbow_angle=160, right_elbow_angle=90,
+            left_knee_angle=90, right_knee_angle=160,
+            feedback=[],
+        )
+        assert detect_diagonal_extension(metrics) == "left_arm_right_leg"
+
+    def test_detect_diagonal_extension_allows_ready_arm_overlap(self):
+        """测试：准备位双臂上举时，单腿对侧伸展仍应识别"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=False, right_leg_extended=True,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=90, right_knee_angle=160,
+            feedback=[],
+        )
+        assert detect_diagonal_extension(metrics) == "left_arm_right_leg"
+
+    def test_detect_diagonal_extension_rejects_both_legs(self):
+        """测试：双腿同时伸展不应识别为可计数对角伸展"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=True, right_leg_extended=True,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=160, right_knee_angle=160,
+            feedback=[],
+        )
+        assert detect_diagonal_extension(metrics) is None
+
+    def test_is_ready_position(self):
+        """测试：膝盖弯曲且四肢收缩应识别为准备姿态"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=False, right_arm_extended=False,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=90, right_elbow_angle=90,
+            left_knee_angle=100, right_knee_angle=100,
+            feedback=[],
+        )
+        assert is_ready_position(metrics)
+
+    def test_is_ready_allows_arms_extended(self):
+        """测试：dead bug 准备位允许双臂上举"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=100, right_knee_angle=100,
+            feedback=[],
+        )
+        assert is_ready_position(metrics)
+
+    def test_is_not_ready_legs_extended(self):
+        """测试：腿部伸展时不识别为准备姿态"""
+        metrics = DeadBugMetrics(
+            left_arm_extended=False, right_arm_extended=False,
+            left_leg_extended=True, right_leg_extended=False,
+            left_elbow_angle=90, right_elbow_angle=90,
+            left_knee_angle=100, right_knee_angle=100,
+            feedback=[],
+        )
+        assert not is_ready_position(metrics)

tests/test_dead_bug_state_machine.py

@@ -0,0 +1,133 @@
+from __future__ import annotations
+
+from app.exercises.dead_bug.state_machine import DeadBugStateMachine
+from app.exercises.dead_bug.types import DeadBugMetrics, DeadBugPhase
+
+class TestDeadBugStateMachine:
+    """死虫式状态机单元测试"""
+
+    def _ready_metrics(self) -> DeadBugMetrics:
+        """构建准备姿态的度量数据"""
+        return DeadBugMetrics(
+            left_arm_extended=False, right_arm_extended=False,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=90, right_elbow_angle=90,
+            left_knee_angle=100, right_knee_angle=100,
+            feedback=[],
+        )
+
+    def _extended_left(self) -> DeadBugMetrics:
+        """构建左臂+右腿对角伸展的度量数据"""
+        return DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=False,
+            left_leg_extended=False, right_leg_extended=True,
+            left_elbow_angle=160, right_elbow_angle=90,
+            left_knee_angle=90, right_knee_angle=160,
+            feedback=[],
+        )
+
+    def _both_legs_extended(self) -> DeadBugMetrics:
+        """构建双腿同时伸展的非标准姿态"""
+        return DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=True, right_leg_extended=True,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=160, right_knee_angle=160,
+            feedback=[],
+        )
+
+    def _arms_extended_ready_legs(self) -> DeadBugMetrics:
+        """构建腿已收回但手臂未收回的姿态"""
+        return DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=100, right_knee_angle=100,
+            feedback=[],
+        )
+
+    def _right_knee_angle(self, angle: float) -> DeadBugMetrics:
+        """构建右膝角连续变化但伸展布尔值尚未稳定的姿态"""
+        return DeadBugMetrics(
+            left_arm_extended=True, right_arm_extended=True,
+            left_leg_extended=False, right_leg_extended=False,
+            left_elbow_angle=160, right_elbow_angle=160,
+            left_knee_angle=90, right_knee_angle=angle,
+            feedback=[],
+        )
+
+    def test_initial_state(self):
+        """测试：状态机初始化后应为READY且计数为0"""
+        sm = DeadBugStateMachine()
+        assert sm.phase == DeadBugPhase.READY
+        assert sm.rep_count == 0
+
+    def test_no_transition_in_ready(self):
+        """测试：准备姿态下不触发状态转换"""
+        sm = DeadBugStateMachine()
+        result = sm.update(self._ready_metrics())
+        assert sm.phase == DeadBugPhase.READY
+        assert result.rep_count == 0
+
+    def test_confirm_extension(self):
+        """测试：连续确认帧数后从READY转换到EXTENDING"""
+        sm = DeadBugStateMachine(extension_confirm_frames=2, reset_confirm_frames=2)
+        sm.update(self._extended_left())
+        assert sm.phase == DeadBugPhase.READY
+        sm.update(self._extended_left())
+        assert sm.phase == DeadBugPhase.EXTENDING
+
+    def test_confirm_extension_from_knee_angle_trend(self):
+        """测试：膝角连续上升时，不依赖单帧伸展布尔值也能确认伸展"""
+        sm = DeadBugStateMachine(extension_confirm_frames=2, reset_confirm_frames=2)
+
+        sm.update(self._right_knee_angle(100))
+        sm.update(self._right_knee_angle(130))
+        assert sm.phase == DeadBugPhase.READY
+        sm.update(self._right_knee_angle(145))
+        sm.update(self._right_knee_angle(150))
+
+        assert sm.phase == DeadBugPhase.EXTENDING
+
+    def test_full_rep_counts_once_after_strict_reset(self):
+        """测试：确认伸展后，只有严格回到准备姿态才计一次"""
+        sm = DeadBugStateMachine(extension_confirm_frames=2, reset_confirm_frames=2)
+
+        sm.update(self._extended_left())
+        sm.update(self._extended_left())
+        assert sm.phase == DeadBugPhase.EXTENDING
+
+        sm.update(self._arms_extended_ready_legs())
+        assert sm.rep_count == 0
+        assert sm.phase == DeadBugPhase.NEED_RESET
+
+        sm.update(self._ready_metrics())
+        result = sm.update(self._ready_metrics())
+        assert result.rep_count == 1
+        assert sm.phase == DeadBugPhase.READY
+
+        result = sm.update(self._ready_metrics())
+        assert result.rep_count == 1
+
+    def test_both_legs_do_not_start_rep(self):
+        """测试：双腿同时伸展不进入计数流程"""
+        sm = DeadBugStateMachine(extension_confirm_frames=2, reset_confirm_frames=2)
+        sm.update(self._both_legs_extended())
+        result = sm.update(self._both_legs_extended())
+
+        assert result.rep_count == 0
+        assert sm.phase == DeadBugPhase.READY
+
+    def test_no_pose_preserves_confirmed_rep_until_reset(self):
+        """测试：已确认伸展后短暂丢姿态，回到准备位仍能完成计数"""
+        sm = DeadBugStateMachine(extension_confirm_frames=2, reset_confirm_frames=2)
+        sm.update(self._extended_left())
+        sm.update(self._extended_left())
+        assert sm.phase == DeadBugPhase.EXTENDING
+
+        sm.mark_no_pose()
+        sm.update(self._ready_metrics())
+        result = sm.update(self._ready_metrics())
+
+        assert result.rep_count == 1
+        assert sm.phase == DeadBugPhase.READY

tests/test_ice_parser.py

@@ -0,0 +1,39 @@
+from __future__ import annotations
+
+from app.signaling.ice_parser import parse_ice
+
+class TestIceParser:
+    """ICE候选者解析单元测试"""
+
+    def test_parse_valid_ice(self):
+        """测试：解析有效的ICE host候选者"""
+        data = {
+            "candidate": "1234567890 1 UDP 2130706431 192.168.1.1 12345 typ host",
+            "sdpMid": "0",
+            "sdpMLineIndex": 0,
+        }
+        cand = parse_ice(data)
+        assert cand is not None
+        assert cand.foundation == "1234567890"
+        assert cand.component == 1
+        assert cand.protocol == "udp"
+        assert cand.ip == "192.168.1.1"
+        assert cand.port == 12345
+        assert cand.type == "host"
+
+    def test_parse_invalid_ice(self):
+        """测试：解析无效ICE字符串应返回None"""
+        assert parse_ice({"candidate": "invalid"}) is None
+
+    def test_parse_srflx(self):
+        """测试：解析含有raddr/rport的srflx候选者"""
+        data = {
+            "candidate": "abcdef 1 UDP 1686052607 203.0.113.1 50000 typ srflx raddr 192.168.1.1 rport 12345",
+            "sdpMid": "0",
+            "sdpMLineIndex": 0,
+        }
+        cand = parse_ice(data)
+        assert cand is not None
+        assert cand.type == "srflx"
+        assert cand.relatedAddress == "192.168.1.1"
+        assert cand.relatedPort == 12345