OpenCV加载YOLO模型实现目标检测

OpenCV加载YOLOv3-tiny模型实现目标检测

YOLOv3-tiny是YOLOv3移动端支持的网络模型版本，是可以在CPU端实时运行的对象检测网络。
首先加载网络，如果需要输出各层信息可将bool变量debug设为true值。

//load model
Net cnn_net = readcnn_netFromDarkcnn_net(yolo_tiny_cfg, yolo_tiny_model);
vector<String> outNames = cnn_net.getUnconnectedOutLayersNames();
if(debug)
	for (int i = 0; i < outNames.size(); i++) 
	{
		cout<<"output layer name : "<<outNames[i].c_str())<<endl;
	}

然后从类名文件中读取类名：

//read classes names
vector<string> classNamesVec;
ifstream classNamesFile("./object_detection_classes_yolov3.txt");
if (classNamesFile.is_open())
{
	string className = "";
	while (etline(classNamesFile, className))
		classNamesVec.push_back(className);
}

因为YOLOv3是多个层的合并输出，所以在OpenCV中调用时候必须显示声明哪些层是输出层，对此OpenCV提供了一个API：

vector<String> cv::dnn::Net::getUnconnectedOutLayersNames()const

来获取所有的输出层名称,该函数返回所有非连接的输出层。

调用时候，必须显式通过输入参数完成推断，相关API如下：

void cv::dnn::Net::forward(
	OutputArrayOfArrays outputBlobs,
	const std::vector< String > & outBlobNames
)

其中outputBlobs是调用之后的输出，outBlobNames是所有输出层的名称。
YOLO的输出前四个为 [center_x, center_y, width, height],后面的是所有类别的得分，这个时候只要得到score最大的对应对象类别，得到检测结果。
完整代码如下：

#include <fstream>
#include <iostream>
#include <algorithm>
#include <cstdlib>
#include <opencv2/opencv.hpp>
#include <opencv2/dnn.hpp>

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

//model path
String yolo_tiny_model = "./yolov3-tiny.weights";
String yolo_tiny_cfg = "./yolov3-tiny.cfg";

const bool debug = true

int main(int argc, char** argv)
{
	//load model
	Net cnn_net = readcnn_netFromDarkcnn_net(yolo_tiny_cfg, yolo_tiny_model);
	vector<String> outNames = cnn_net.getUnconnectedOutLayersNames();
	if(debug)
		for (int i = 0; i < outNames.size(); i++) 
		{
			cout<<"output layer name : "<<outNames[i].c_str())<<endl;
		}
	
	//read classes names
	vector<string> classNamesVec;
	ifstream classNamesFile("./object_detection_classes_yolov3.txt");
	if (classNamesFile.is_open())
	{
		string className = "";
		while (etline(classNamesFile, className))
			classNamesVec.push_back(className);
	}

	//load image 
	Mat input_image = imread("./test.jpg");
	Mat inputBlob = blobFromImage(input_image, 1 / 255.F, Size(416, 416), Scalar(), true, false);
	cnn_net.setInput(inputBlob);

	// detection
	ector<Mat> outs;
	cnn_net.forward(outs, outNames);
	vector<double> layersTimings;
	double freq = getTickFrequency() / 1000;
	double time = cnn_net.getPerfProfile(layersTimings) / freq;
	vector<Rect> boxes;
	vector<int> classIds;
	vector<float> confidences;
	for (size_t i = 0; i<outs.size(); ++i)
	{
		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;
			minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
			if (confidence > 0.5)
			{
				int centerX = (int)(data[0] * input_image.cols);
				int centerY = (int)(data[1] * input_image.rows);
				int width = (int)(data[2] * input_image.cols);
				int height = (int)(data[3] * input_image.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));
			}
		}
	}

	vector<int> indices;
	NMSBoxes(boxes, confidences, 0.5, 0.2, indices);
	for (size_t i = 0; i < indices.size(); ++i)
	{
		int idx = indices[i];
		Rect box = boxes[idx];
		String className = classNamesVec[classIds[idx]];
		putText(input_image, className.c_str(), box.tl(), FONT_HERSHEY_SIMPLEX, 1.0, Scalar(255, 0, 0), 2, 8);
		rectangle(input_image, box, Scalar(0, 0, 255), 2, 8, 0);
	}

	imshow("detections", input_image);
	waitKey(0);
	return 0;	
}

本文代码需在OpenCV4.0以上版本运行。