jira:NYJ-1550 desc:m3u8 images

+import base64
+import hashlib
+import subprocess
+from pathlib import Path
+
+import cv2
+
+try:
+    from .m3u8_to_mp4_sei import (
+        download_m3u8_to_mp4,
+        extract_h264_es_from_mp4,
+        parse_h264_sei,
+        extract_utc_from_sei,
+    )
+except ImportError:
+    from m3u8_to_mp4_sei import (
+        download_m3u8_to_mp4,
+        extract_h264_es_from_mp4,
+        parse_h264_sei,
+        extract_utc_from_sei,
+    )
+
+# 用于区分传入的 start_utc/end_utc 是绝对 UTC 时间戳还是秒偏移
+_IS_ABSOLUTE_UTC_THRESHOLD = 1_000_000_000
+
+
+class Video2Frame:
+    def __init__(self, cache_dir):
+        self.cache_dir = Path(cache_dir)
+        self.video_dir = self.cache_dir / "videos"
+        self.frames_dir = self.cache_dir / "frames"
+        self.video_dir.mkdir(parents=True, exist_ok=True)
+        self.frames_dir.mkdir(parents=True, exist_ok=True)
+
+    def to_frames(self, url, start_utc, end_utc, fps, roi=None, max_px_area=None) -> list:
+        """
+        下载 m3u8 视频流的指定片段并提取帧，保存到 cache_dir 下。
+
+        - video_dir 中保存的是 start_utc ~ end_utc 截取后的视频片段（而非完整视频）。
+        - 如果视频中包含 SEI UTC 信息且传入的 start_utc/end_utc 为绝对时间戳，
+          则会基于 SEI UTC 进行帧级定位；否则将 start_utc/end_utc 视为秒偏移量。
+
+        Args:
+            url: m3u8 视频流地址。
+            start_utc: 截取开始时间（秒偏移或绝对 UTC 时间戳）。
+            end_utc: 截取结束时间（秒偏移或绝对 UTC 时间戳）。
+            fps: 目标抽帧帧率。
+            roi: 感兴趣区域 (x, y, w, h)，可选。
+            max_px_area: 最大像素面积，超过则等比例缩小，可选。
+
+        Returns:
+            提取的帧图片路径列表。
+        """
+        if end_utc <= start_utc:
+            raise ValueError("end_utc 必须大于 start_utc")
+
+        unique_str = f"{url}_{start_utc}_{end_utc}"
+        unique_id = hashlib.md5(unique_str.encode("utf-8")).hexdigest()
+
+        clip_path = self.video_dir / f"{unique_id}.mp4"
+        frame_output_dir = self.frames_dir / unique_id
+        frame_output_dir.mkdir(parents=True, exist_ok=True)
+
+        # 判断是否为绝对 UTC 时间戳
+        is_absolute_utc = (
+            start_utc > _IS_ABSOLUTE_UTC_THRESHOLD
+            and end_utc > _IS_ABSOLUTE_UTC_THRESHOLD
+        )
+
+        # 若片段未缓存，先截取目标片段再保存
+        if not clip_path.exists():
+            if is_absolute_utc:
+                # 绝对 UTC 模式：先下载完整视频到临时文件，解析 SEI 后截取片段
+                temp_path = self.video_dir / f"{unique_id}_full.mp4"
+                download_m3u8_to_mp4(url, str(temp_path))
+
+                # 提取 SEI UTC 信息
+                utc_records = []
+                try:
+                    es_data = extract_h264_es_from_mp4(str(temp_path))
+                    sei_list = parse_h264_sei(es_data)
+                    utc_records = extract_utc_from_sei(sei_list)
+                except Exception:
+                    pass
+
+                if utc_records:
+                    utc_list = [r["utc"] for r in utc_records]
+                    start_frame_idx = self._find_frame_idx(
+                        utc_list, start_utc, find_last_not_greater=False
+                    )
+                    end_frame_idx = self._find_frame_idx(
+                        utc_list, end_utc, find_last_not_greater=True
+                    )
+
+                    cap_temp = cv2.VideoCapture(str(temp_path))
+                    video_fps = cap_temp.get(cv2.CAP_PROP_FPS)
+                    cap_temp.release()
+                    if video_fps <= 0:
+                        video_fps = 25.0
+
+                    start_sec = start_frame_idx / video_fps
+                    duration = (end_frame_idx - start_frame_idx) / video_fps
+                    self._ffmpeg_extract_clip(
+                        str(temp_path), str(clip_path), start_sec, duration
+                    )
+                else:
+                    # 无 SEI 无法定位，直接将完整视频重命名为片段
+                    temp_path.rename(clip_path)
+
+                # 清理临时完整视频
+                if temp_path.exists():
+                    temp_path.unlink(missing_ok=True)
+            else:
+                # 秒偏移模式：直接用 ffmpeg 从 URL 截取片段
+                duration = end_utc - start_utc
+                self._ffmpeg_download_clip(
+                    url, str(clip_path), start_utc, duration
+                )
+
+        # 从截取后的片段中提取帧
+        cap = cv2.VideoCapture(str(clip_path))
+        if not cap.isOpened():
+            raise ValueError(f"无法打开视频片段: {clip_path}")
+
+        # total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
+        video_fps = cap.get(cv2.CAP_PROP_FPS)
+        if video_fps <= 0:
+            video_fps = 25.0
+
+        # 计算抽帧间隔
+        frame_interval = int(video_fps / fps) if fps > 0 else int(video_fps)
+        if frame_interval < 1:
+            frame_interval = 1
+
+        frames = []
+        saved_count = 0
+        frame_count = 0
+        while True:
+            ret, frame = cap.read()
+            if not ret:
+                break
+
+            if frame_count % frame_interval == 0:
+                if roi is not None:
+                    x, y, w, h = roi
+                    frame = frame[y:y + h, x:x + w]
+                frame = self.reisize_frame(frame, max_px_area)
+                frames.append(frame)
+                frame_path = frame_output_dir / f"frame_{saved_count:06d}.jpg"
+                cv2.imwrite(str(frame_path), frame)
+                saved_count += 1
+            frame_count += 1
+        cap.release()
+        return frames
+
+    def reisize_frame(self, frame, max_px_area):
+        if max_px_area is None:
+            return frame
+        h, w = frame.shape[:2]
+        area = h * w
+        if area > max_px_area:
+            scale = (max_px_area / area) ** 0.5
+            new_w = int(w * scale)
+            new_h = int(h * scale)
+            frame = cv2.resize(frame, (new_w, new_h), interpolation=cv2.INTER_AREA)
+        return frame
+
+    @staticmethod
+    def _find_frame_idx(utc_list, target_utc, find_last_not_greater=True):
+        """
+        在 utc_list 中查找最接近 target_utc 的帧索引。
+
+        Args:
+            utc_list: 按帧顺序排列的 UTC 列表。
+            target_utc: 目标 UTC。
+            find_last_not_greater: True 返回最后一个 <= target_utc 的索引；
+                                   False 返回第一个 >= target_utc 的索引。
+        """
+        if not utc_list:
+            return 0
+
+        if find_last_not_greater:
+            idx = 0
+            for i, utc in enumerate(utc_list):
+                if utc > target_utc:
+                    break
+                idx = i
+            return idx
+        else:
+            for i, utc in enumerate(utc_list):
+                if utc >= target_utc:
+                    return i
+            return len(utc_list) - 1
+
+    @staticmethod
+    def _ffmpeg_extract_clip(input_path, output_path, start_sec, duration):
+        """使用 ffmpeg 从本地视频截取指定片段。"""
+        cmd = [
+            "ffmpeg", "-y",
+            "-i", input_path,
+            "-ss", str(start_sec),
+            "-t", str(duration),
+            "-c", "copy",
+            "-bsf:a", "aac_adtstoasc",
+            "-movflags", "+faststart",
+            output_path,
+        ]
+        try:
+            subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        except subprocess.CalledProcessError as e:
+            raise RuntimeError(
+                f"ffmpeg 截取片段失败: {e.stderr.decode('utf-8', errors='ignore')}"
+            )
+
+    @staticmethod
+    def _ffmpeg_download_clip(url, output_path, start_sec, duration):
+        """使用 ffmpeg 从 URL 下载并截取指定片段。"""
+        cmd = [
+            "ffmpeg", "-y",
+            "-ss", str(start_sec),
+            "-fflags", "+discardcorrupt",
+            "-i", url,
+            "-t", str(duration),
+            "-c", "copy",
+            "-bsf:a", "aac_adtstoasc",
+            "-movflags", "+faststart",
+            output_path,
+        ]
+        try:
+            subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+        except subprocess.CalledProcessError as e:
+            raise RuntimeError(
+                f"ffmpeg 下载片段失败: {e.stderr.decode('utf-8', errors='ignore')}"
+            )
+
+
+def frames2content(frames):
+    """
+    将帧列表（图片路径列表）转为 LLM content 列表。
+    格式与 llm_video_content.py 中的 contents 返回的 content 一致。
+    """
+    contents = []
+    video_prompt = (
+        f"以下是从视频中按时间顺序提取的 {len(frames)} 帧画面，"
+        f"请将它们视为一个连续的视频进行分析。"
+    )
+    contents.append({"type": "text", "text": video_prompt})
+
+    for frame in frames:
+        _, buffer = cv2.imencode('.jpg', frame)
+        b64_str = base64.b64encode(buffer).decode('utf-8')
+
+        # with open(frame_path, "rb") as f:
+        #     img_bytes = f.read()
+        # b64_str = base64.b64encode(img_bytes).decode("utf-8")
+
+        contents.append({
+            "type": "image_url",
+            "image_url": {
+                "url": f"data:image/jpeg;base64,{b64_str}"
+            }
+        })
+
+    return contents
+
+
+if __name__ == "__main__":
+    # pass
+
+    import argparse
+    import sys
+    import tempfile
+
+    parser = argparse.ArgumentParser(description="Video2Frame 测试脚本")
+    parser.add_argument(
+        "--url",
+        default=r"D:\pythonProject\learn\3b3e99cf4ca84c3782503d8817242de2.mp4",
+        help="测试用 m3u8 地址（默认使用公开测试流）",
+    )
+    parser.add_argument(
+        "--cache-dir",
+        default=r"C:\Users\lzw\AppData\Local\Temp\video2frame_test_mn26tcy_my",
+        help="缓存目录（默认自动创建临时目录）",
+    )
+    parser.add_argument(
+        "--start",
+        type=float,
+        default=0,
+        help="截取开始时间，单位：秒（默认 0）",
+    )
+    parser.add_argument(
+        "--end",
+        type=float,
+        default=10,
+        help="截取结束时间，单位：秒（默认 10）",
+    )
+    parser.add_argument(
+        "--fps",
+        type=float,
+        default=2.0,
+        help="抽帧帧率（默认 1.0）",
+    )
+    args = parser.parse_args()
+
+    cache_dir = args.cache_dir or tempfile.mkdtemp(prefix="video2frame_test_")
+    print(f"[INFO] 缓存目录: {cache_dir}")
+
+    v2f = Video2Frame(cache_dir)
+
+    # TEST 1: 基本抽帧
+    print(
+        f"\n[TEST 1] to_frames: url={args.url}, "
+        f"start={args.start}s, end={args.end}s, fps={args.fps}"
+    )
+    try:
+        frames = v2f.to_frames(
+            url=args.url,
+            start_utc=args.start,
+            end_utc=args.end,
+            fps=args.fps,
+            roi=None,
+            max_px_area=200_000,
+        )
+        print(f"  成功提取 {len(frames)} 帧")
+        if frames:
+            print(f"  首帧: {frames[0].shape}")
+            print(f"  末帧: {frames[-1].shape}")
+    except Exception as e:
+        print(f"  失败: {e}")
+        import traceback
+
+        traceback.print_exc()
+        sys.exit(1)
+
+    # TEST 2: frames2content
+    print("\n[TEST 2] frames2content")
+    content = frames2content(frames)
+    print(f"  content 列表长度: {len(content)}")
+    for item in content[:3]:
+        preview = (
+            item["text"][:60] + "..."
+            if item["type"] == "text"
+            else item["image_url"]["url"][:60] + "..."
+        )
+        print(f"  - type={item['type']}: {preview}")
+    if len(content) > 3:
+        print(f"  ... 还有 {len(content) - 3} 个元素")
+
+    print("\n[TEST 3] FootballReplayVideoEvent 进球识别（可选）")
+
+    try:
+        from ..util.football_replay_video_event_by_llm import FootballReplayVideoEvent
+    except ImportError:
+        import sys
+        from pathlib import Path
+        sys.path.insert(0, str(Path(__file__).parent.parent))
+        from util.football_replay_video_event_by_llm import FootballReplayVideoEvent
+
+    fbrv = FootballReplayVideoEvent(
+        base_url="http://192.168.1.59:11434",
+        model="qwen3.6:35b-a3b-q8_0",
+        temperature=0.7,
+    )
+    event_json = fbrv.video_event_for_contents(content, None)
+    print(f"  识别结果: {event_json}")
+
+
+    # # TEST 3: ROI + max_px_area
+    # print("\n[TEST 3] to_frames with roi + max_px_area")
+    # try:
+    #     end_crop = min(args.start + 5, args.end)
+    #     frames_cropped = v2f.to_frames(
+    #         url=args.url,
+    #         start_utc=args.start,
+    #         end_utc=end_crop,
+    #         fps=1.0,
+    #         roi=None,
+    #         max_px_area=200_000,
+    #     )
+    #     print(f"  成功提取 {len(frames_cropped)} 帧（带 ROI 裁剪和面积缩放）")
+    # except Exception as e:
+    #     print(f"  跳过/失败: {e}")
+    #
+    # # TEST 4: SEI UTC 模式说明（仅演示，需真实含 SEI 的流才能运行）
+    # print("\n[TEST 4] SEI UTC 绝对时间戳模式（示例代码，需替换为含 SEI 的流）")
+    # print(
+    #     "  # 示例：假设视频第一帧 UTC 为 1715600000，提取 10 秒片段\n"
+    #     "  # frames = v2f.to_frames(url, start_utc=1715600000, end_utc=1715600010, fps=1)"
+    # )
+    #
+    # print("\n[INFO] 所有测试完成")