开发环境

    VS2013 + OpenCV3.4.1 + Qt5.8.0

实验准备

    VGG_VOC0712_SSD_300x300_iter_120000.caffemodel

    deploy.prototxt

    以上数据可以由这里下载得到(或者直接下载本工程)，使用的是基于Caffe+VOC0712数据集训练出来的caffemodel。

主要代码

读取网络

string modelTxt = "caffe_ssd_300x300/deploy.prototxt";
string modelBin = "caffe_ssd_300x300/VGG_VOC0712_SSD_300x300_iter_120000.caffemodel";
try{

	net = dnn::readNetFromCaffe(modelTxt, modelBin);
}
catch (cv::Exception &ee){

	QMessageBox::warning(this, "Exception", ee.what());
	if (net.empty()){

		QMessageBox::warning(this, "Exception", "Can't load the network by using the flowing files.");
		return;
	}
}

前向识别

Mat frame;
image.copyTo(frame);
if (frame.empty()){

	QMessageBox::warning(this, "Warning", "image is empty, please check!");
	return;
}
if (frame.channels() == 4) cvtColor(frame, frame, COLOR_BGRA2BGR);
	
double ttt = (double)cvGetTickCount();

Mat inputBlob = blobFromImage(frame, 1.0f, Size(300, 300), Scalar(104, 117, 123), false, false); //Convert Mat to batch of images
net.setInput(inputBlob, "data");						//set the network input
Mat detection = net.forward("detection_out");			//compute output

ostringstream ss;
vector<double> layersTimings;
Mat detectionMat(detection.size[2], detection.size[3], CV_32F, detection.ptr<float>());

ttt = (double)cvGetTickCount() - ttt;
ui.labelTime->setText(toChinese("识别时间：") + QString::number(ttt / (cvGetTickFrequency() * 1000000)) + toChinese("秒"));

float confidenceThreshold = ui.dsbConfidence->value();
for (int i = 0; i < detectionMat.rows; i++){

	float confidence = detectionMat.at<float>(i, 2);

	if (confidence > confidenceThreshold){

		size_t objectClass = (size_t)(detectionMat.at<float>(i, 1));
		int xLeftBottom = static_cast<int>(detectionMat.at<float>(i, 3) * frame.cols);
		int yLeftBottom = static_cast<int>(detectionMat.at<float>(i, 4) * frame.rows);
		int xRightTop	= static_cast<int>(detectionMat.at<float>(i, 5) * frame.cols);
		int yRightTop	= static_cast<int>(detectionMat.at<float>(i, 6) * frame.rows);
		ss.str("");
		ss << confidence;
		String conf(ss.str());
		Rect object(xLeftBottom, yLeftBottom, xRightTop - xLeftBottom, yRightTop - yLeftBottom);
		rectangle(frame, object, Scalar(0, 0, 255));
		//String label = String(classNames[objectClass]) + ": " + conf;
		String label = String(classNamesZH[objectClass]) + ": " + conf;
		
		int baseLine = 0;
		Size labelSize = getTextSize(label, FONT_HERSHEY_SIMPLEX, 0.48, 1, &baseLine);
		rectangle(frame, Rect(Point(xLeftBottom, yLeftBottom + labelSize.height / 2 - 4),
			Size(labelSize.width, labelSize.height + baseLine)),
			Scalar(255, 255, 255), FILLED);

		//putText(frame, label, Point(xLeftBottom, yLeftBottom), FONT_HERSHEY_SIMPLEX, 0.48, Scalar(0, 0, 0));
		putTextZH(frame, label.c_str(), Point(xLeftBottom, yLeftBottom), Scalar(0, 0, 255), 14, "Arial");
	}
}

Mat detectionMat是输入图像后经过网络前向传播后的输出7*10的结果矩阵，其定义如下图所示：

目标为20分类，定义如下：

const char* classNames[] = { "background",
"aeroplane", "bicycle", "bird", "boat",
"bottle", "bus", "car", "cat", "chair",
"cow", "diningtable", "dog", "horse",
"motorbike", "person", "pottedplant",
"sheep", "sofa", "train", "tvmonitor" };

const char* classNamesZH[] = { 
"背景","飞机", "自行车", "鸟", "船",
"瓶子", "巴士", "汽车", "猫", "椅子",
"牛", "餐桌", "狗", "马","摩托车", 
"人", "盆栽","羊", "沙发", "火车", "电视" };

上图中置信概率最高(0.999)的目标数组下标为2，对应的是bicycle自行车，只要大于设置的阈值(变量confidenceThreshold)，就会在图像上标记出目标的位置(detectionMat行向量的3,4,5,6元素)。比如将阈值confidenceThreshold设置为0.5，则识别结果只有一个是大于0.5的，则只会在图像上标记出自行车，如下图：

实验效果

本地图片

相机实时

在相机上(普通的USB相机)会比较卡顿···，因为用的是CPU跑的，附上本人帅照···

附件

    源代码工程戳这里(注：release下的可执行程序可以直接运行)。