OpenCV的人脸检测模型FaceDetectorYN

1. 官网地址

https://docs.opencv.org/4.x/df/d20/classcv_1_1FaceDetectorYN.html
FaceDetectorYN是opencv内置的一个人脸检测方法，利用的是yunet。
这是一个DNN-based face detector.模型的下载地址：
https://github.com/opencv/opencv_zoo/tree/master/models/face_detection_yunet

2. 怎样利用

2.1.到opencv_zoo下载模型文件和代码

2.2. 下载文件展示

2.3. 修改了demo支持读取视频文件，默认是图片和摄像头

1. # This file is part of OpenCV Zoo project.
2. # It is subject to the license terms in the LICENSE file found in the same directory.
3. #
4. # Copyright (C) 2021, Shenzhen Institute of Artificial Intelligence and Robotics for Society, all rights reserved.
5. # Third party copyrights are property of their respective owners.
7. import argparse
9. import numpy as np
10. import cv2 as cv
12. # Check OpenCV version
13. opencv_python_version = lambda str_version: tuple(map(int, (str_version.split("."))))
14. assert opencv_python_version(cv.__version__) >= opencv_python_version("4.10.0"), \
15.     "Please install latest opencv-python for benchmark: python3 -m pip install --upgrade opencv-python"
17. from yunet import YuNet
19. # Valid combinations of backends and targets
20. backend_target_pairs = [
21.     [cv.dnn.DNN_BACKEND_OPENCV, cv.dnn.DNN_TARGET_CPU],
22.     [cv.dnn.DNN_BACKEND_CUDA, cv.dnn.DNN_TARGET_CUDA],
23.     [cv.dnn.DNN_BACKEND_CUDA, cv.dnn.DNN_TARGET_CUDA_FP16],
24.     [cv.dnn.DNN_BACKEND_TIMVX, cv.dnn.DNN_TARGET_NPU],
25.     [cv.dnn.DNN_BACKEND_CANN, cv.dnn.DNN_TARGET_NPU]
26. ]
28. parser = argparse.ArgumentParser(
29.     description='YuNet: A Fast and Accurate CNN-based Face Detector (https://github.com/ShiqiYu/libfacedetection).')
30. parser.add_argument('--input', '-i', type=str,
31.                     help='Usage: Set input to a certain image, omit if using camera.')
32. parser.add_argument('--model', '-m', type=str, default='face_detection_yunet_2023mar.onnx',
33.                     help="Usage: Set model type, defaults to 'face_detection_yunet_2023mar.onnx'.")
34. parser.add_argument('--backend_target', '-bt', type=int, default=0,
35.                     help='''Choose one of the backend-target pair to run this demo:
36.                         {:d}: (default) OpenCV implementation + CPU,
37.                         {:d}: CUDA + GPU (CUDA),
38.                         {:d}: CUDA + GPU (CUDA FP16),
39.                         {:d}: TIM-VX + NPU,
40.                         {:d}: CANN + NPU
41.                     '''.format(*[x for x in range(len(backend_target_pairs))]))
42. parser.add_argument('--conf_threshold', type=float, default=0.7,
43.                     help='Usage: Set the minimum needed confidence for the model to identify a face, defauts to 0.9. Smaller values may result in faster detection, but will limit accuracy. Filter out faces of confidence < conf_threshold.')
44. parser.add_argument('--nms_threshold', type=float, default=0.3,
45.                     help='Usage: Suppress bounding boxes of iou >= nms_threshold. Default = 0.3.')
46. parser.add_argument('--top_k', type=int, default=5000,
47.                     help='Usage: Keep top_k bounding boxes before NMS.')
48. parser.add_argument('--save', '-s', action='store_true',
49.                     help='Usage: Specify to save file with results (i.e. bounding box, confidence level). Invalid in case of camera input.')
50. parser.add_argument('--vis', '-v', action='store_true',
51.                     help='Usage: Specify to open a new window to show results. Invalid in case of camera input.')
52. parser.add_argument('--camera_or_video', '-c', default='123.mov',
53.                     help='Usage: Specify to open camera or video')
54. args = parser.parse_args()
57. def visualize(image, results, box_color=(0, 255, 0), text_color=(0, 255, 255), fps=None):
58.     output = image.copy()
59.     landmark_color = [
60.         (0, 0, 255),  # right eye
61.         (0, 0, 255),  # left eye
62.         (0, 255, 0),  # nose tip
63.         (255, 0, 255),  # right mouth corner
64.         (0, 255, 255)  # left mouth corner
65.     ]
67.     if fps is not None:
68.         cv.putText(output, 'FPS: {:.2f}'.format(fps), (50, 50), cv.FONT_HERSHEY_SIMPLEX, 1.5, text_color)
70.     for det in results:
71.         bbox = det[0:4].astype(np.int32)
72.         cv.rectangle(output, (bbox[0], bbox[1]), (bbox[0] + bbox[2], bbox[1] + bbox[3]), box_color, 2)
74.         conf = det[-1]
75.         cv.putText(output, '{:.4f}'.format(conf), (bbox[0], bbox[1] + bbox[3] // 2), cv.FONT_HERSHEY_DUPLEX, 1.5, text_color)
77.         landmarks = det[4:14].astype(np.int32).reshape((5, 2))
78.         for idx, landmark in enumerate(landmarks):
79.             cv.circle(output, landmark, 4, landmark_color[idx], 10)
81.     return output
84. if __name__ == '__main__':
85.     backend_id = backend_target_pairs[args.backend_target][0]
86.     target_id = backend_target_pairs[args.backend_target][1]
88.     # Instantiate YuNet
89.     model = YuNet(modelPath=args.model,
90.                   inputSize=[320, 320],
91.                   confThreshold=args.conf_threshold,
92.                   nmsThreshold=args.nms_threshold,
93.                   topK=args.top_k,
94.                   backendId=backend_id,
95.                   targetId=target_id)
97.     # If input is an image
98.     if args.input is not None:
99.         image = cv.imread(args.input)
100.         h, w, _ = image.shape
102.         # Inference
103.         model.setInputSize([w, h])
104.         results = model.infer(image)
106.         # Print results
107.         print('{} faces detected.'.format(results.shape[0]))
108.         for idx, det in enumerate(results):
109.             print(
110.                 '{}: {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f} {:.0f}'.format(
111.                     idx, *det[:-1])
112.             )
114.         # Draw results on the input image
115.         image = visualize(image, results)
117.         # Save results if save is true
118.         if args.save:
119.             print('Resutls saved to result.jpg\n')
120.             cv.imwrite('result.jpg', image)
122.         # Visualize results in a new window
123.         if args.vis:
124.             cv.namedWindow(args.input, cv.WINDOW_AUTOSIZE)
125.             cv.imshow(args.input, image)
126.             cv.waitKey(0)
127.     else:  # Omit input to call default camera
128.         deviceId = args.camera_or_video
129.         cap = cv.VideoCapture(int(deviceId) if deviceId.isdigit() else deviceId)
130.         w = int(cap.get(cv.CAP_PROP_FRAME_WIDTH))
131.         h = int(cap.get(cv.CAP_PROP_FRAME_HEIGHT))
132.         model.setInputSize([w, h])
133.         fps = int(cap.get(cv.CAP_PROP_FPS))
135.         # 定义视频编码器和创建VideoWriter对象
136.         fourcc = cv.VideoWriter_fourcc(*'mp4v')  # 或者使用 'XVID'
137.         out = cv.VideoWriter('output.mp4', fourcc, fps, (w, h))
140.         tm = cv.TickMeter()
141.         while cv.waitKey(1) < 0:
142.             # Inference
143.             tm.start()
145.             hasFrame, frame = cap.read()
146.             if not hasFrame:
147.                 print('No frames grabbed!')
148.                 break
150.             results = model.infer(frame)  # results is a tuple
151.             tm.stop()
153.             # Draw results on the input image
154.             frame = visualize(frame, results, fps=tm.getFPS())
156.             # Visualize results in a new Window
157.             cv.imshow('YuNet face detection', frame)
158.             # tm.reset()
159.             # 写入视频文件
160.             out.write(frame)
162.         out.release()
163.         cap.release()
165. cv.destroyAllWindows()

复制代码

## 2.4 效果展示


     吕一_faces_detection
  

免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！更多信息从访问主页：qidao123.com:ToB企服之家，中国第一个企服评测及商务社交产业平台。