Abstract: The visual odometry plays a key role in the positioning and navigation of mobile robots, but the current algorithm still has room for improvement in terms of running speed, trajectory accuracy, and robustness. In order to improve the accuracy of camera trajectory, a loop-closure detection algorithm is proposed based on the sparse direct method. The algorithm directly extracts two features to form a hybrid feature point to improve the robustness of the system. The hybrid feature point is used to track and match key frames, so that the visual odometry can detect closed loops, and then the pose map is used to optimize the positioning accuracy. Experimental results show strong robustness in a complex environment and a balance between speed and accuracy.

Key words: visual odometry, loop-closure detection, sparse direct method, visual feature, camera trajectory