2つの画像のキーポイントをマッチする既存ロジック→このベクトル探索を置き換えて、低速マッチングを作成する。

opencv_matchpoint_sample.cpp

Vector<KeyPoint> keyPoint0 = keypoints_vec[0];
Vector<KeyPoint> keyPoint1 = keypoints_vec[1];
Mat descriptor0 = descriptors_vec[0];
Mat descriptor1 = descriptors_vec[1];

vector<DMatch> matches;
BFMatcher matcher(NORM_HAMMING, true);

matcher.match(descriptor0, descriptor1, matches);

double max_dist = 0;
double min_dist = DBL_MAX;
for (int j = 0; j < (int) matches.size(); j++) {
  double dist = matches[j].distance;
	if (dist < min_dist)
		min_dist = dist;
	if (dist > max_dist)
		max_dist = dist;
}

// 良いペアのみ残す
double cutoff = 3.0 * (min_dist + 1.0);
std::set<int> existing_trainIdx;
std::vector<cv::DMatch> matches_good;
for (int j = 0; j < (int) matches.size(); j++) {
	if (matches[j].trainIdx <= 0)
		matches[j].trainIdx = matches[j].imgIdx;
	if (matches[j].distance > 0.0 && matches[j].distance < cutoff) {
		if (existing_trainIdx.find(matches[j].trainIdx)
				== existing_trainIdx.end() && matches[j].trainIdx >= 0
				&& matches[j].trainIdx < (int) keyPoint1.size()) {
			matches_good.push_back(matches[j]);
			existing_trainIdx.insert(matches[j].trainIdx);
		}
	}
}