python读取宇视摄像头实时流，并用opencv做人脸检测

爱被打了一巴掌 2021-09-19 15:00 1763阅读 0赞

一、测试摄像头

宇视摄像头默认密码123456，取流地址：rtsp://admin:123456@192.168.1.134:554

下载VLC播放器进行验证，下载地址如下

[https://dl.pconline.com.cn/html\_2/1/124/id=6661&pn=0&linkPage=1.html][https_dl.pconline.com.cn_html_2_1_124_id_6661_pn_0_linkPage_1.html]

选择媒体-打开网络串流

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70][]

可以看到预览流。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70 1][]

二、准备python 库

1、下载cv2

安装cv2模块不能直接用命令pip install cv2安装，要用pip install opencv-python安装，

如果失败，可以下载whl:[https://www.lfd.uci.edu/~gohlke/pythonlibs/\#opencv][https_www.lfd.uci.edu_gohlke_pythonlibs_opencv]

或者[https://pypi.tuna.tsinghua.edu.cn/simple/opencv-python/][https_pypi.tuna.tsinghua.edu.cn_simple_opencv-python]

一定要下载和python版本相适应的，比如我的3.6必须下

*  opencv\_python‑3.4.6‑cp36‑cp36m‑win\_amd64.whl，下cp37都会提示is not a supported wheel on this platform

pip install opencv\_python‑3.4.6‑cp36‑cp36m‑win\_amd64.whl

三、利用opencv预览

#以下是最常用的读取视频流的方法
    import cv2
    url = 'rtsp://admin:123456@192.168.1.134:554'
    cap = cv2.VideoCapture(url)
    while(cap.isOpened()):
        ret, frame = cap.read()
    # Display the resulting frame  
        cv2.imshow('frame',frame)
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    # When everything done, release the capture  
    cap.release()
    cv2.destroyAllWindows()

可以看到跳出预览画面：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70 2][]

四、人脸检测

分类器可以用opencv自带的haarcascade_frontalface_default.xml，在安装路径下：\Lib\site-packages\cv2\data\haarcascade_frontalface_default.xml，要和代码放在同一文件夹下
    
    代码如下：
    import cv2
    def CatVideo():
        #1调用摄像头，0表示从pc本地连接的摄像头，url表示从ipc的url取流
        cap=cv2.VideoCapture(url)
        classfier = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
        color = (0, 255, 0)
        while cap.isOpened():
            ok,frame=cap.read()
            if not ok:
                break
                # 3灰度转换
            grey = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
            # 4人脸检测，1.2和2分别为图片缩放比例和需要检测的有效点数
            faceRects = classfier.detectMultiScale(grey, scaleFactor=1.2, minNeighbors=3, minSize=(32, 32))
            if len(faceRects) > 0:  # 大于0则检测到人脸
                print("检测到人脸")
                for faceRect in faceRects:  # 单独框出每一张人脸
                    x, y, w, h = faceRect  # 5画图
                    cv2.rectangle(frame, (x - 10, y - 10), (x + w + 10, y + h + 10), color, 3)
            cv2.imshow("Face Recognition", frame)
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
        cap.release()
        cv2.destroyAllWindows()

其中：waitKey（）方法本身表示等待键盘输入，

参数是1，表示延时1ms切换到下一帧图像，对于视频而言；

参数为0，如cv2.waitKey(0)只显示当前帧图像，相当于视频暂停,；

按q时退出播放

![20190613164622771.png][]

五、问题

1、误报率还是挺高的，

2、解码经常出错

\[h264 @ 000000000ef76940\] cabac decode of qscale diff failed at 84 17  
\[h264 @ 000000000ef76940\] error while decoding MB 84 17, bytestream 507ffmpeg

解码h264流出错，最挫的方法是设置IPC编码格式

3、启动较慢

[https_dl.pconline.com.cn_html_2_1_124_id_6661_pn_0_linkPage_1.html]: https://dl.pconline.com.cn/html_2/1/124/id=6661&pn=0&linkPage=1.html
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70]: /images/20210726/6a219ff13b40420d8b36d23b1fbebed4.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70 1]: /images/20210726/4223c0e86343462cab812ddaf3d2d122.png
[https_www.lfd.uci.edu_gohlke_pythonlibs_opencv]: https://www.lfd.uci.edu/~gohlke/pythonlibs/#opencv
[https_pypi.tuna.tsinghua.edu.cn_simple_opencv-python]: https://pypi.tuna.tsinghua.edu.cn/simple/opencv-python/
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2x1YW5zag_size_16_color_FFFFFF_t_70 2]: /images/20210726/b519c53c05b141e2b854d48f8af8bac7.png
[20190613164622771.png]: /images/20210726/9f957518cf92406484d76c46b960c848.png