ImageRecognition/CharacterDectect.cpp

#include "CharacterDectect.h"
using namespace std;

bool CharacterDectect::Init(bool isCuda)
{
	string model_path = "models/character-sim.onnx";
	try {
		net = cv::dnn::readNet(model_path);
	}
	catch (const std::exception& ex)
	{
		YunDaISASImageRecognitionService::ConsoleLog(ex.what());
		return false;
	}
	//cuda
	if (isCuda) {
		net.setPreferableBackend(cv::dnn::DNN_BACKEND_CUDA);
		net.setPreferableTarget(cv::dnn::DNN_TARGET_CUDA_FP16);
	}
	//cpu
	else {
		net.setPreferableBackend(cv::dnn::DNN_BACKEND_DEFAULT);
		net.setPreferableTarget(cv::dnn::DNN_TARGET_CPU);
	}
	return true;

}
IDetection::DectectResult CharacterDectect::GetStateResult(cv::Mat img, cv::Rect rec)
{
	//resultValue.clear();
	std::cout << "test" << std::endl;
	try
	{
		cv::Mat ROI = img(rec);
		/*imwrite("test.png", ROI);
		YunDaISASImageRecognitionService::SetImage(QString::fromStdString("test.png"));*/
		Detect(ROI);
	}
	catch (const std::exception& ex)
	{
		YunDaISASImageRecognitionService::ConsoleLog(ex.what());
	}
	if (resultValue.m_confidence < 0.1)
	{
		resultValue = DectectResult(0.45, 0, className[1]);
	}
	return resultValue;
}
IDetection::DectectResult CharacterDectect::GetDigitResult(cv::Mat img, cv::Rect rec)
{
	return resultValue;
}
vector<IDetection::DectectResult> CharacterDectect::GetStateResults()
{
	return resultStateValues;
}
vector<IDetection::DectectResult> CharacterDectect::GetDigitResults()
{
	return resultDigitValues;
}
bool CharacterDectect::Detect(cv::Mat& SrcImg) {
	cv::Mat blob;
	int col = SrcImg.cols;
	int row = SrcImg.rows;
	int maxLen = MAX(col, row);
	cv::Mat netInputImg = SrcImg.clone();
	if (maxLen > 1.2 * col || maxLen > 1.2 * row) {
		cv::Mat resizeImg = cv::Mat::zeros(maxLen, maxLen, CV_8UC3);
		SrcImg.copyTo(resizeImg(cv::Rect(0, 0, col, row)));
		netInputImg = resizeImg;
	}
	cv::dnn::blobFromImage(netInputImg, blob, 1 / 255.0, cv::Size(netWidth, netHeight), cv::Scalar(0, 0, 0), true, false);
	net.setInput(blob);
	std::vector<cv::Mat> netOutputImg;
	net.forward(netOutputImg, net.getUnconnectedOutLayersNames());
	std::vector<int> classIds;//结果id数组
	std::vector<float> confidences;//结果每个id对应置信度数组
	std::vector<cv::Rect> boxes;//每个id矩形框
	float ratio_h = (float)netInputImg.rows / netHeight;
	float ratio_w = (float)netInputImg.cols / netWidth;
	int net_width = className.size() + 5;  //输出的网络宽度是类别数+5
	float* pdata = (float*)netOutputImg[0].data;
	for (int stride = 0; stride < strideSize; stride++) {    //stride
		int grid_x = (int)(netWidth / netStride[stride]);
		int grid_y = (int)(netHeight / netStride[stride]);
		for (int anchor = 0; anchor < 3; anchor++) {	//anchors
			const float anchor_w = netAnchors[stride][anchor * 2];
			const float anchor_h = netAnchors[stride][anchor * 2 + 1];
			for (int i = 0; i < grid_y; i++) {
				for (int j = 0; j < grid_x; j++) {
					float box_score = pdata[4]; ;//获取每一行的box框中含有某个物体的概率
					if (box_score >= boxThreshold) {
						cv::Mat scores(1, className.size(), CV_32FC1, pdata + 5);
						cv::Point classIdPoint;
						double max_class_socre;
						minMaxLoc(scores, 0, &max_class_socre, 0, &classIdPoint);
						max_class_socre = (float)max_class_socre;
						if (max_class_socre >= classThreshold)
						{
							//rect [x,y,w,h]
							float x = pdata[0];  //x
							float y = pdata[1];  //y
							float w = pdata[2];  //w
							float h = pdata[3];  //h
							int left = (x - 0.5 * w) * ratio_w;
							int top = (y - 0.5 * h) * ratio_h;

							int widthBox = int(w * ratio_w);
							int heightBox = int(h * ratio_h);
							widthBox = widthBox > col ? col : widthBox;
							heightBox = heightBox > row ? row : heightBox;
							left = left < 0 ? 0 : left;
							top = top < 0 ? 0 : top;
							if (left < 0 || left>col || top < 0 || top>row || widthBox > col || heightBox > row)
							{
								continue;
							}
							classIds.push_back(classIdPoint.x);
							confidences.push_back(max_class_socre * box_score);
							boxes.push_back(cv::Rect(left, top, widthBox, heightBox));
						}
					}
					pdata += net_width;//下一行
				}
			}
		}
	}

	//执行非最大抑制以消除具有较低置信度的冗余重叠框（NMS）
	vector<int> nms_result;
	cv::dnn::NMSBoxes(boxes, confidences, nmsScoreThreshold, nmsThreshold, nms_result);
	float confidenceMax = 0;
	int confidenceMaxId = 0;
	vector<pair<int, float>> classidConfidence;
	resultStateValues.clear();
	
	if (nms_result.size() > 0)
	{
		for (int i = 0; i < nms_result.size(); i++) {
			int idx = nms_result[i];
			if (confidences[idx] > confidenceMax)
			{
				confidenceMax = confidences[idx];

				resultValue = DectectResult(confidenceMax, 0, className[classIds[idx]]);
			}
			YunDaISASImageRecognitionService::ConsoleLog(QString::fromStdString(className[classIds[idx]]));
			auto item = std::find_if(classidConfidence.begin(), classidConfidence.end(), [idx](pair<int, float> p) { return p.first == idx; });
			if (item != classidConfidence.end())
			{
				if (item->second< confidences[idx])
				{
					item->second = confidences[idx];
				}
			}
			else {
				classidConfidence.push_back(pair<int, float>(idx, confidences[idx]));
			}
		}
		if (classidConfidence.size()>0)
		{
			for (size_t i = 0; i < classidConfidence.size(); i++)
			{
				if (classidConfidence[i].second>0.65)
				{
					resultStateValues.push_back(DectectResult(classidConfidence[i].second, 0, className[classIds[classidConfidence[i].first]]));
				}
			}
		}
	}
	else
	{
		resultValue = DectectResult(confidenceMax, 0, className[1]);
	}

	return true;

}