百度飞浆之PaddleHub视频移动人脸识别

百度AI Studio原项目链接------点击此处前往

安装paddlehub

使用pip简单操作即可 pip install paddlehub

paddlehub丰富的预训练模型链接：点击此处前往

步入正题

导入所需包

import cv2
import paddlehub as hub
import numpy as np

调用paddlehub人脸检测模块，并将处理结果以帧的形式保存至result.mp4，其中face_detection_result文件夹保存了每一帧的人脸识别结果。

# 定义检测人脸函数，返回识别的图片路径
def face_det(img):
    face_detector = hub.Module(name="pyramidbox_lite_server")
    result = face_detector.face_detection(images=[img], visualization='true', output_dir='face_detection_result')
    return result[0]['path']

# 定义视频处理函数
def video_pro(video_path, save_path):
    cap = cv2.VideoCapture(video_path)
    fourcc = cv2.VideoWriter_fourcc(*'XVID')
    fps = cap.get(cv2.CAP_PROP_FPS)
    frame_count = cap.get(cv2.CAP_PROP_FRAME_COUNT)
    print('帧率：%d'%fps)
    print('帧数：%d'%frame_count)
    size = (int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)),int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)))
    out = cv2.VideoWriter(save_path, fourcc, fps, size)
    count = 1
    while cap.isOpened():
        ret, frame = cap.read()
        if ret:
            result_path = 'face_detection_result' + '/' + face_det(frame).split('.')[0] + '.jpg'
            old_img = cv2.imread(result_path)
            # 处理保存的图片，旋转
            new_img = np.rot90(old_img)
            new_img = np.rot90(new_img)
            new_img = np.rot90(new_img)
            out.write(new_img)
            k = cv2.waitKey(1) & 0xFF
            if k == 27:
                break
            elif count == frame_count:
                break
            else:
                count += 1
        else:
            break
        
    cap.release()
    out.release()
    cv2.destroyAllWindows()

最后调用上述函数即可。

video_pro('test.mp4', 'result.avi')
cap.release()
out.release()
cv2.destroyAllWindows()

处理结果请点击链接观看效果：点击此处前往