PokerNet-poker recognition: 基于人工神经网络的扑克识别 （5）

结果解读

• 结果1
• 结果2

结果1

void computeBCValue(cv::Mat mat_img, std::vector<bbox_t> result_vec, std::vector<std::string> obj_names,
	int current_det_fps = -1, int current_cap_fps = -1)
{
	int const colors[6][3] = { { 1,0,1 },{ 0,0,1 },{ 0,1,1 },{ 0,1,0 },{ 1,1,0 },{ 1,0,0 } };
	Cardsets cardsets;
	int tempx = 0;
	vector<bbox_t>  mypoints;    //

	for (auto &i : result_vec) {
		if (i.prob < confidencetheta) { // if I am not confident than 50%, skip this poker
			continue;
		}

		// here we compute the distance w.r.t (0,0)
		if (i.h < i.w) {// 3rd
			if (i.prob > cardsets.confidence[2]) {
				cardsets.confidence[2] = i.prob;
				cardsets.cardvalues[2] = Point3f(i.x, i.y, i.obj_id);
			}

		}
		else {
			mypoints.push_back(i);
		}

	}
	if (cardsets.cardvalues[2].x > 0) {
		//cout << "3rd-" << obj_names[cardsets.cardvalues[2].z] << endl;

	}

	if (mypoints.size() < 2) { return; }

	sort(mypoints.begin(), mypoints.end(), SetSortRule);

	int first = 0;
	int second = 0;

	if (mypoints.size() == 4) {// 

		if (mypoints[0].prob > mypoints[1].prob) {
			//cout << "1st-" << obj_names[mypoints[0].obj_id] << endl;
			first = 0;
		}
		else {
			//cout << "1st-" << obj_names[mypoints[1].obj_id] << endl;
			first = 1;
		}

		if (mypoints[2].prob > mypoints[3].prob) {
			//	cout << "2ed-" << obj_names[mypoints[2].obj_id] << endl;
			second = 2;
		}
		else {
			//cout << "2ed-" << obj_names[mypoints[3].obj_id] << endl;
			second = 3;
		}

	}
	else if (mypoints.size() == 3) {// if we only detect 3 points, take first one to match with other two decide which is pair

		/* details of the theroy are on the notebook page   */

		int delta_y = mypoints[1].y - mypoints[0].y;

		if (getCosineDistance(mypoints[0], mypoints[1]) < getCosineDistance(mypoints[0], mypoints[2]) && delta_y > 100) { // we assume they belong to one card
			first = mypoints[0].prob > mypoints[1].prob ? 0 : 1;
			second = 2;

		}
		else {// 0 and 1 not one card
			//	cout << "1st-" << obj_names[mypoints[0].obj_id] << endl;
			second = mypoints[1].prob > mypoints[2].prob ? 1 : 2;

		}

	}
	else  if (mypoints.size() == 2) {

		if (getCosineDistance(mypoints[0], mypoints[1]) < cosinetheta) { // we assume they belong to one card
			first = mypoints[0].prob > mypoints[1].prob ? 0 : 1;

		}
		else {

			second = 1;
		}

	}

	cardsets.confidence[0] = mypoints[first].prob;
	cardsets.cardvalues[0] = Point3f(mypoints[first].x, mypoints[first].y, mypoints[first].obj_id);
	cardsets.confidence[1] = mypoints[second].prob;
	cardsets.cardvalues[1] = Point3f(mypoints[second].x, mypoints[second].y, mypoints[second].obj_id);
	ostringstream info;

	info << "1st-" << obj_names[mypoints[first].obj_id] << ",confidence-" << setprecision(2) << fixed << mypoints[first].prob << ",";
	info << "2ed-" << obj_names[mypoints[second].obj_id] << ",confidence-" << setprecision(2) << fixed << mypoints[second].prob << ",";

	if (cardsets.cardvalues[2].x > 0) { info << "3rd-" << obj_names[cardsets.cardvalues[2].z] << ",confidence-" << setprecision(2) << fixed << cardsets.confidence[2] << ","; }
	cout << info.str() << endl;
	putText(mat_img, info.str(), cv::Point(5, 40), cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 255));

}

结果2

void computeDTValue(cv::Mat mat_img, std::vector<bbox_t> result_vec, std::vector<std::string> obj_names,
	int current_det_fps = -1, int current_cap_fps = -1)
{
	int const colors[6][3] = { { 1,0,1 },{ 0,0,1 },{ 0,1,1 },{ 0,1,0 },{ 1,1,0 },{ 1,0,0 } };
	int tempx = 0;
	vector<bbox_t>  mypoints;    //

	for (auto &i : result_vec) {
		if (i.prob > confidencetheta) { // if I am not confident than 50%, skip this poker
			mypoints.push_back(i);
		}
	}

	sort(mypoints.begin(), mypoints.end(), SetSortRulebyConfidence);

	int cardids = 0;
	float confidence = 0;
	if (mypoints.size() > 0) {
		cardids = mypoints[0].obj_id;
		confidence = mypoints[0].prob;

	}
	else { //error

		cout << "nothing detected" << endl;
	}

	ostringstream info;

	info << "1st-" << obj_names[cardids] << ",confidence-" << setprecision(2) << fixed << confidence << ",";
	cout << info.str() << endl;
	cv::putText(mat_img, info.str(), cv::Point(5, 40), cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 255));

}