opencv DNN模块之人脸检测

DNN人脸检测

在3.x版本人连检测，常用haar级联器检测，受光照、人脸位置影响比较大，识别率也不理想，DNN模块给出专用人脸模型，跟踪效果很好，抗干扰能力很强。在4版本之后DNN检测以后将会成为主流。

---

opencv提供的模型

• 基于SSD网络模型caffe训练生成的人脸检测模型
• 基于SSD网络模型tensflow训练生成的人脸检测模型
• opencv模型量化版本16FP/uint8

res10_300x300_ssd_iter_140000_fp16.caffemodel
deploy.prototxt
opencv_face_detector_uint8.pb
opencv_face_detector.pbtxt

---

相关参数

# OpenCV's face detection network
opencv_fd:
  model: "opencv_face_detector.caffemodel"
  config: "opencv_face_detector.prototxt"
  mean: [104, 177, 123]
  scale: 1.0
  width: 300
  height: 300
  rgb: false
  sample: "object_detection"

---

以caffe模型为例的代码演示

#include 
#include 
#include 

using namespace std;
using namespace cv;
using namespace cv::dnn;

int main(void)
{
    string bin_model = "/work/opencv_dnn/face_detector/res10_300x300_ssd_iter_140000_fp16.caffemodel";
    string protxt = "/work/opencv_dnn/face_detector/deploy.prototxt";
    // load network model
    Net net = readNetFromCaffe(protxt, bin_model);

    // 设置计算后台
    net.setPreferableBackend(DNN_BACKEND_OPENCV);
    net.setPreferableTarget(DNN_TARGET_CPU);
    namedWindow("检测画面",WINDOW_AUTOSIZE);

    // 获取各层信息
    vector<string> layer_names = net.getLayerNames();
    for (int  i = 0; i < layer_names.size(); i++) {
        int id = net.getLayerId(layer_names[i]);
        auto layer = net.getLayer(id);
        printf("layer id : %d, type : %s, name : %s \n", id, layer->type.c_str(), layer->name.c_str());
    }
    VideoCapture capture(0);
    Mat frame;
    while (true) {
        bool ret = capture.read(frame);
        if (!ret) break;
        flip(frame,frame,1);
        // 构建输入
        Mat blob = blobFromImage(frame, 1, Size(300, 300), Scalar(104, 177, 123), false, false);
        net.setInput(blob, "data");
        // 执行推理
        Mat detection = net.forward("detection_out");
        Mat detectionMat(detection.size[2], detection.size[3], CV_32F, detection.ptr<float>());
        float confidence_threshold = 0.5;
        // 解析输出数据
        for (int i = 0; i < detectionMat.rows; i++) {
            float* curr_row = detectionMat.ptr<float>(i);
            int image_id = (int)(*curr_row++);
            size_t objIndex = (size_t)(*curr_row++);
            float score = *curr_row++;
            if (score > confidence_threshold) {
                float tl_x = (*curr_row++) * frame.cols;
                float tl_y = (*curr_row++) * frame.rows;
                float br_x = (*curr_row++) * frame.cols;
                float br_y = (*curr_row++) * frame.rows;
                Rect box((int)tl_x, (int)tl_y, (int)(br_x - tl_x), (int)(br_y - tl_y));
                rectangle(frame, box, Scalar(0, 0, 255), 2, 8, 0);
                putText(frame, "man face", box.tl(), FONT_HERSHEY_SIMPLEX, 1.5, Scalar(0, 255, 0), 3, 8);
            }
        }
        // measure time consume
        vector<double> layersTimings;
        double freq = getTickFrequency() / 1000.0;
        double time = net.getPerfProfile(layersTimings) / freq;
        ostringstream ss;
        ss << "FPS: " << 1000 / time << " ; time : " << time << " ms";

        // show
        putText(frame, ss.str(), Point(20, 20), FONT_HERSHEY_PLAIN, 1.0, Scalar(255, 0, 0), 2, 8);
        imshow("检测画面", frame);
        char c = waitKey(1);
        if (c == 27) { // ESC
            break;
        }
    }

    // 释放资源
    capture.release();
    waitKey(0);
    destroyAllWindows();
    return 0;
}

---

效果

---