1 CascadeClassifier 级联分类器人脸识别

有两种：haar级联和lbp级联，我用brew安装的，级联文件在/opt/homebrew/Cellar/opencv/4.5.5_2/share/opencv4/haarcascades里面，haar级联文件大小是900kb左右，lbp级联文件大小是50kb左右。
检测前需要将图像转化成灰度图，并做直方图均衡化处理。

lbp的文件大小、识别速度和效果都要好于haar。

总的来说，人脸检测效果很一般，人脸不动时检测框会闪烁，人脸稍有偏转或遮挡就检测不到。

int myFaceDetect(int argc, char** argv) {
    double w = 0, h = 0, fps = 24;
    Mat frame;
    Mat gray;
    Mat res;
    VideoCapture cap;
    
    if (!cap.open(0)) {
        return 0;
    }
    w = cap.get(CAP_PROP_FRAME_WIDTH);
    h = cap.get(CAP_PROP_FRAME_HEIGHT);
    
    printf("cap w: %f, h: %f\n", w, h);
    
    namedWindow("cam");
    while(true) {
        auto tick = getTickCount();
//        cap >> frame;
        cap.read(frame);
        if (frame.empty()) {
            break;
        }
        flip(frame, frame, 1);
        cvtColor(frame, gray, COLOR_BGRA2GRAY);
        equalizeHist(gray, gray);
        
//        auto ccPath = "/opt/homebrew/Cellar/opencv/4.5.5_2/share/opencv4/haarcascades/haarcascade_frontalface_extended.xml";
        auto ccPath = "/opt/homebrew/Cellar/opencv/4.5.5_2/share/opencv4/lbpcascades/lbpcascade_frontalface_improved.xml";
        CascadeClassifier cc;
        if (!cc.load(ccPath)) {
            cout << "load CascadeClassifier failed" << endl;
            return -1;
        }
        vector<Rect> faces;
        cc.detectMultiScale(gray, faces);
        for (int i = 0; i < faces.size(); i++) {
            rectangle(frame, faces[i], Scalar(0, 0, 255));
        }
        imshow("cam", frame);
        auto time = (getTickCount() - tick) / getTickFrequency();
        printf("handleTime: %f\n", time);
        
        if (waitKey(1000 / fps) == ' ') {
            break;
        }
    }
    destroyAllWindows();
    return 0;
}

2 DNN 深度神经网络人脸识别

需要下载神经网络模型和描述文件，模型大小为2.7Mb，描述文件大小为35kb

OpenCV的dnn支持caffe和TensorFlow两种模型，我这里用的是TensorFlow的模型。

检测直接用原始图像就行。

人脸检测效果非常好，人脸偏转或者遮挡一半仍能检测到。缺点是计算时间长一点，在移动端会明显一点。

int myDnnFaceDetect(int argc, char** argv) {
    double w = 0, h = 0, fps = 24;
    Mat frame;
    Mat gray;
    Mat res;
    VideoCapture cap;
    
    if (!cap.open(0)) {
        return 0;
    }
    w = cap.get(CAP_PROP_FRAME_WIDTH);
    h = cap.get(CAP_PROP_FRAME_HEIGHT);
    
    printf("cap w: %f, h: %f\n", w, h);
    
    auto pb_path = "/Users/chenrongchao/Downloads/face_detector-main/opencv_face_detector_uint8.pb";
    auto pbtext_path = "/Users/chenrongchao/Downloads/face_detector-main/opencv_face_detector.pbtxt";
    dnn::Net net = dnn::readNetFromTensorflow(pb_path, pbtext_path);
    
    namedWindow("cam");
    while(true) {
        auto tick = getTickCount();
//        cap >> frame;
        cap.read(frame);
        if (frame.empty()) {
            break;
        }
        flip(frame, frame, 1);
//        cvtColor(frame, gray, COLOR_BGRA2GRAY);
        
        auto blob = dnn::blobFromImage(frame, 1.0, Size2i(300, 300), Scalar(104,177,123),false,false);
        net.setInput(blob);
        auto probs = net.forward();
        Mat detectionMat(probs.size[2], probs.size[3], CV_32F, probs.ptr<float>());
        //解析结果
        for (int i = 0; i < detectionMat.rows; i++) {
            float confidence = detectionMat.at<float>(i, 2);
            if (confidence > 0.5) { //提取矩形四个角的坐标
                int x1 = static_cast<int>(detectionMat.at<float>(i, 3)*frame.cols);
                int y1 = static_cast<int>(detectionMat.at<float>(i, 4)*frame.rows);
                int x2 = static_cast<int>(detectionMat.at<float>(i, 5)*frame.cols);
                int y2 = static_cast<int>(detectionMat.at<float>(i, 6)*frame.rows);
                Rect box(x1, y1, x2 - x1, y2 - y1); //红色矩形框
                rectangle(frame, box, Scalar(0, 0, 255), 4, 8, 0); //标记人脸
            }
        }
        imshow("cam", frame);
        auto time = (getTickCount() - tick) / getTickFrequency();
        printf("handleTime: %f\n", time);
        
        if (waitKey(1000 / fps) == ' ') {
            break;
        }
    }
    destroyAllWindows();
    return 0;
}