I have been studying DLoopDetector for the past two days, and I want to use the ORB feature extraction method for loop detection. However, the original demo uses the feature extraction of BRIEF256, and the extracted descriptor is stored in a vector, which is completely different from the Mat data type of the descriptor that uses the ORB method for feature extraction. Therefore, many codes do not It is not fully compatible. It is necessary to convert the Mat type descriptor into a vector<cv::Mat> type for storage. It took me a lot of brain cells to adapt it from the day before yesterday to the basic realization today. Moreover, I didn't move from 3:00 to 5:30 pm this afternoon. I concentrated on looking at the code. I found the corresponding solution through the demo of DBoW2.
The specific results are shown below. There are certain errors and they have not been fully realized, but generally speaking, they are mostly correct.
It is generally correct, and most of it is correct compared with before, and there are still some unmatched content. I will continue to study.