opencv4+yolov3部署目标检测

opencv4+yolov3在windows上部署的代码如下：

// This code is written at BigVision LLC. It is based on the OpenCV project.
//It is subject to the license terms in the LICENSE file found in this distribution and at http://opencv.org/license.html

// Usage example:  ./object_detection_yolo.out --video=run.mp4
//                 ./object_detection_yolo.out --image=bird.jpg
#include <fstream>
#include <sstream>
#include <iostream>

#include <opencv2/dnn.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/highgui.hpp>
;
using namespace cv;
using namespace dnn;
using namespace std;

string pro_dir = "F:/wangjiao/result/weights/"; 

                                                  // Initialize the parameters
float confThreshold = 0.5; // Confidence threshold
float nmsThreshold = 0.4;  // Non-maximum suppression threshold
int inpWidth = 416;  // Width of network's input image
int inpHeight = 416; // Height of network's input image
vector<string> classes;

// Remove the bounding boxes with low confidence using non-maxima suppression
void postprocess(Mat& frame, const vector<Mat>& out);

// Draw the predicted bounding box
void drawPred(int classId, float conf, int left, int top, int right, int bottom, Mat& frame);

// Get the names of the output layers
vector<String> getOutputsNames(const Net& net);

void detect_image(string image_path, string modelWeights, string modelConfiguration, string classesFile);

void detect_video(string video_path, string modelWeights, string modelConfiguration, string classesFile);

int main(int argc, char** argv)
{
    // Give the configuration and weight files for the model
    String modelConfiguration = pro_dir + "default/models/yolov3.cfg";
    String modelWeights = pro_dir + "default/models/yolov3.weights";
    string image_path = pro_dir + "default/images/dog.jpg";
    string classesFile = pro_dir + "default/models/coco.names";// "coco.names";
    detect_image(image_path, modelWeights, modelConfiguration, classesFile);
    //string video_path = pro_dir + "data/images/run.mp4";
    //detect_video(video_path, modelWeights, modelConfiguration, classesFile);
    cv::waitKey(0);
    return 0;
}

void detect_image(string image_path, string modelWeights, string modelConfiguration, string classesFile) {
    // Load names of classes
    ifstream ifs(classesFile.c_str());
    string line;
    while (getline(ifs, line)) classes.push_back(line);

    // Load the network
    Net net = readNetFromDarknet(modelConfiguration, modelWeights);
    net.setPreferableBackend(DNN_BACKEND_OPENCV);
    net.setPreferableTarget(DNN_TARGET_OPENCL);

    // Open a video file or an image file or a camera stream.
    string str, outputFile;
    cv::Mat frame = cv::imread(image_path);
    // Create a window
    static const string kWinName = "Deep learning object detection in OpenCV";
    namedWindow(kWinName, WINDOW_NORMAL);

    // Stop the program if reached end of video
    // Create a 4D blob from a frame.
    Mat blob;
    blobFromImage(frame, blob, 1 / 255.0, Size(inpWidth, inpHeight), Scalar(0, 0, 0), true, false);

    //Sets the input to the network
    net.setInput(blob);

    // Runs the forward pass to get output of the output layers
    vector<Mat> outs;
    net.forward(outs, getOutputsNames(net));

    // Remove the bounding boxes with low confidence
    postprocess(frame, outs);
    // Put efficiency information. The function getPerfProfile returns the overall time for inference(t) and the timings for each of the layers(in layersTimes)
    vector<double> layersTimes;
    double freq = getTickFrequency() / 1000;
    double t = net.getPerfProfile(layersTimes) / freq;
    string label = format("Inference time for a frame : %.2f ms", t);
    putText(frame, label, Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 0, 255));
    // Write the frame with the detection boxes
    imshow(kWinName, frame);
    cv::waitKey(30);
}

void detect_video(string video_path, string modelWeights, string modelConfiguration, string classesFile) {
    string outputFile = "./yolo_out_cpp.avi";;
    // Load names of classes
    ifstream ifs(classesFile.c_str());
    string line;
    while (getline(ifs, line)) classes.push_back(line);

    // Load the network
    Net net = readNetFromDarknet(modelConfiguration, modelWeights);
    net.setPreferableBackend(DNN_BACKEND_OPENCV);
    net.setPreferableTarget(DNN_TARGET_CPU);

    // Open a video file or an image file or a camera stream.
    VideoCapture cap;
    //VideoWriter video;
    Mat frame, blob;

    try {
        // Open the video file
        ifstream ifile(video_path);
        if (!ifile) throw("error");
        cap.open(video_path);
    }
    catch (...) {
        cout << "Could not open the input image/video stream" << endl;
        return;
    }

    // Get the video writer initialized to save the output video
    //video.open(outputFile, 
    //    VideoWriter::fourcc('M', 'J', 'P', 'G'), 
    //    28, 
    //    Size(cap.get(CAP_PROP_FRAME_WIDTH), cap.get(CAP_PROP_FRAME_HEIGHT)));

    // Create a window
    static const string kWinName = "Deep learning object detection in OpenCV";
    namedWindow(kWinName, WINDOW_NORMAL);

    // Process frames.
    while (waitKey(1) < 0)
    {
        // get frame from the video
        cap >> frame;

        // Stop the program if reached end of video
        if (frame.empty()) {
            cout << "Done processing !!!" << endl;
            cout << "Output file is stored as " << outputFile << endl;
            waitKey(3000);
            break;
        }
        // Create a 4D blob from a frame.
        blobFromImage(frame, blob, 1 / 255.0, Size(inpWidth, inpHeight), Scalar(0, 0, 0), true, false);

        //Sets the input to the network
        net.setInput(blob);

        // Runs the forward pass to get output of the output layers
        vector<Mat> outs;
        net.forward(outs, getOutputsNames(net));

        // Remove the bounding boxes with low confidence
        postprocess(frame, outs);

        // Put efficiency information. The function getPerfProfile returns the overall time for inference(t) and the timings for each of the layers(in layersTimes)
        vector<double> layersTimes;
        double freq = getTickFrequency() / 1000;
        double t = net.getPerfProfile(layersTimes) / freq;
        string label = format("Inference time for a frame : %.2f ms", t);
        putText(frame, label, Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, Scalar(0, 0, 255));

        // Write the frame with the detection boxes
        Mat detectedFrame;
        frame.convertTo(detectedFrame, CV_8U);
        //video.write(detectedFrame);
        imshow(kWinName, frame);

    }

    cap.release();
    //video.release();

}

// Remove the bounding boxes with low confidence using non-maxima suppression
void postprocess(Mat& frame, const vector<Mat>& outs)
{
    vector<int> classIds;
    vector<float> confidences;
    vector<Rect> boxes;

    for (size_t i = 0; i < outs.size(); ++i)
    {
        // Scan through all the bounding boxes output from the network and keep only the
        // ones with high confidence scores. Assign the box's class label as the class
        // with the highest score for the box.
        float* data = (float*)outs[i].data;
        for (int j = 0; j < outs[i].rows; ++j, data += outs[i].cols)
        {
            Mat scores = outs[i].row(j).colRange(5, outs[i].cols);
            Point classIdPoint;
            double confidence;
            // Get the value and location of the maximum score
            minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
            if (confidence > confThreshold)
            {
                int centerX = (int)(data[0] * frame.cols);
                int centerY = (int)(data[1] * frame.rows);
                int width = (int)(data[2] * frame.cols);
                int height = (int)(data[3] * frame.rows);
                int left = centerX - width / 2;
                int top = centerY - height / 2;

                classIds.push_back(classIdPoint.x);
                confidences.push_back((float)confidence);
                boxes.push_back(Rect(left, top, width, height));
            }
        }
    }

    // Perform non maximum suppression to eliminate redundant overlapping boxes with
    // lower confidences
    vector<int> indices;
    NMSBoxes(boxes, confidences, confThreshold, nmsThreshold, indices);
    for (size_t i = 0; i < indices.size(); ++i)
    {
        int idx = indices[i];
        Rect box = boxes[idx];
        drawPred(classIds[idx], confidences[idx], box.x, box.y,
            box.x + box.width, box.y + box.height, frame);
    }
}

// Draw the predicted bounding box
void drawPred(int classId, float conf, int left, int top, int right, int bottom, Mat& frame)
{
    //Draw a rectangle displaying the bounding box
    rectangle(frame, Point(left, top), Point(right, bottom), Scalar(255, 178, 50), 3);

    //Get the label for the class name and its confidence
    string label = format("%.2f", conf);
    if (!classes.empty())
    {
        CV_Assert(classId < (int)classes.size());
        label = classes[classId] + ":" + label;
    }

    //Display the label at the top of the bounding box
    int baseLine;
    Size labelSize = getTextSize(label, FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
    top = max(top, labelSize.height);
    rectangle(frame, Point(left, top - round(1.5*labelSize.height)), Point(left + round(1.5*labelSize.width), top + baseLine), Scalar(255, 255, 255), FILLED);
    putText(frame, label, Point(left, top), FONT_HERSHEY_SIMPLEX, 0.75, Scalar(0, 0, 0), 1);
}

// Get the names of the output layers
vector<String> getOutputsNames(const Net& net)
{
    static vector<String> names;
    if (names.empty())
    {
        //Get the indices of the output layers, i.e. the layers with unconnected outputs
        vector<int> outLayers = net.getUnconnectedOutLayers();

        //get the names of all the layers in the network
        vector<String> layersNames = net.getLayerNames();

        // Get the names of the output layers in names
        names.resize(outLayers.size());
        for (size_t i = 0; i < outLayers.size(); ++i)
            names[i] = layersNames[outLayers[i] - 1];
    }
    return names;
}

参考文章：

https://github.com/PanJinquan/opencv-learning-tutorials/blob/master/dnn_tutorial/object_detection_yolov3_image.cpp（即该项目的object_detection_yolov3_image.cpp）

注意：需要把cvSize(inpWidth, inpHeight)改为Size(inpWidth, inpHeight),[版本问题]