Learnning Dlib(六) Speeding up Dlib’s Facial Landmark Detector

原文
通过图片的实验！Dlib 无法达到移动端视频的要求！但是可以根据上文进行优化,原理很简单，就是把图片缩小0.4,取得坐标，然后在放大 4 倍，就是原图的坐标，这时速度还是可以在移动端应用。

这篇博客比较详细
用openCv 测试了下 速度提升上去了，但是精确度却大打折扣。

还有一个方案就是用系统的AVMetadataFaceObject，去脸的区域，然后在用dlib 进行人脸对齐，取点。
相关demo
GitHub

openCv 对齐相关代码

- (void)openCv{

    NSLog(@"openCv Start");
    //get image path
    std::string fileName =[imagePath UTF8String];

     NSString *harr = [[NSBundle mainBundle] pathForResource:@"haarcascade_frontalface_alt2" ofType:@"xml"];
     std::string harrName =[harr UTF8String];

    cv::CascadeClassifier faceDetector;
    faceDetector.load(harrName);



    // Change to dlib's image format. No memory is copied.
    array2d<rgb_pixel> img;
    array2d<rgb_pixel> img_small;

    //load ios image
    load_image(img,fileName);
    cv::Mat face = cv::imread(fileName);
//    cv::resize(img, img_small, cv::Size(), 1.0/FACE_DOWNSAMPLE_RATIO, 1.0/FACE_DOWNSAMPLE_RATIO);

    std::vector<cv::Rect> faces;
    cv::Mat face_gray;
    cvtColor( face, face_gray, CV_BGR2GRAY);  //rgb类型转换为灰度类型
    equalizeHist(face_gray, face_gray);   //直方图均衡化
    faceDetector.detectMultiScale(face_gray,faces,1.1,3,0|CV_HAAR_DO_CANNY_PRUNING,cv::Size(30,30));

    // faceDetector.detectMultiScale(face_gray, faces, 1.1, 3, 0);  // 分类器对象调用
    NSLog(@"检测到人脸个数：%lu\n", faces.size());
    printf("检测到人脸个数：%lu\n", faces.size());
    if (faces.size() == 0) {
        return;
    }


     for (int i = 0;i < faces.size(); i++) {
        dlib::rectangle det;
        //将opencv检测到的矩形转换为dlib需要的数据结构，这里没有判断检测不到人脸的情况
        det.set_left(faces[i].x);
        det.set_top(faces[i].y);
        det.set_right(faces[i].x+faces[i].width);
        det.set_bottom(faces[i].y+faces[i].height);

        std::vector<full_object_detection> shapes;
        full_object_detection shape = sp(img, det);

        // and draw them into the image (samplebuffer)
        for (unsigned long k = 0; k < shape.num_parts(); k++) {
            point p = shape.part(k);
//            NSLog(@"point %lu  frmae %ld",k,p.x());

            // p 点的直径 3 参数为原点直径 rgb_pixel 颜色
            draw_solid_circle(img, p, 2, dlib::rgb_pixel(0, 255, 255));

//            std::cout << "number of parts: "<< shape.num_parts() << std::endl;
//            std::cout << "pixel position of first part:  " << shape.part(0) << std::endl;
//            std::cout << "pixel position of second part: " << shape.part(1) <<  std::endl;
            // You get the idea, you can get all the face part locations if
            // you want them.  Here we just store them in shapes so we can
            // put them on the screen.
            shapes.push_back(shape);
        }

    }

    NSArray *docPath = NSSearchPathForDirectoriesInDomains(NSDocumentDirectory,NSUserDomainMask, YES);
    NSString *documentsPath = [docPath objectAtIndex:0];
    documentsPath = [documentsPath stringByAppendingPathComponent:@"test.png"];
    const char *savePath = [documentsPath UTF8String];
    save_jpeg(img, savePath);
    self.imageView.image = [UIImage imageWithContentsOfFile:documentsPath];

    NSLog(@"openCv End");

    cout << "Hit enter to process the next image..." << endl;
    cin.get();

}