经过优化后，稳定在60ms，不卡顿

import cv2 from ultralytics import solutions import torch # ✅ 必须在文件顶部 新增导入torch！！！ import gc # ✅ 必须在文件顶部 新增导入gc！！！ def open_iphone_camera_with_cv(): """ 主函数：找到并打开iPhone摄像头 """ print("正在检测iPhone摄像头...\n") # for i in range(5): cap = cv2.VideoCapture(0) if cap.isOpened(): print("打开默认摄像头 (索引0)") # 强制开启硬件加速解码（M1 Metal） cap.set(cv2.CAP_PROP_HW_ACCELERATION, cv2.VIDEO_ACCELERATION_ANY) # 开启帧缓冲区优化，降低延迟 cap.set(cv2.CAP_PROP_BUFFERSIZE, 1) cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640) cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 480) region_points = [(20, 400), (1080, 400), (1080, 360), (20, 360)] # w, h, fps = (int(cap.get(x)) for x in (cv2.CAP_PROP_FRAME_WIDTH, cv2.CAP_PROP_FRAME_HEIGHT, cv2.CAP_PROP_FPS)) # video_writer = cv2.VideoWriter("object_counting_output.avi", cv2.VideoWriter_fourcc(*"mp4v"), fps, (w, h)) counter = solutions.ObjectCounter( show=False, # display the output region=region_points, # pass region points model="yolo26x.pt", # model="yolo26n-obb.pt" for object counting with OBB model. device="mps", half=False, conf=0.3, # 置信度调高一点，减少无效检测，加速推理 iou=0.45, # NMS IOU阈值 max_det=50, # 每张图像的最大检测数量 verbose=True, # 是否打印详细信息 # save_results=True, # 是否保存结果到文件 # classes=[0, 2], # count specific classes, e.g., person and car with the COCO pretrained model. tracker="botsort.yaml", # choose trackers, e.g., "bytetrack.yaml" ) while True: ret, frame = cap.read() if ret: frame_corrected = cv2.flip(frame, 1) results = counter(frame_corrected) # 显存+内存清理，根治耗时上涨 try: torch.mps.empty_cache() except Exception: pass gc.collect() cv2.imshow(f'Camera M2 Pro | 推理≈13ms 跟踪≈9ms | 按q退出', results.plot_im) if cv2.waitKey(1) & 0xFF == ord('q'): break cv2.destroyAllWindows() cap.release() print("摄像头已关闭") else: print("无法打开摄像头，请检查连接。") # 运行 if __name__ == "__main__": # 方法1：自动检测并打开iPhone摄像头 open_iphone_camera_with_cv()