Line Segment Semi-direct Tracking Based on Stereo Odometry

The feature based visual SLAM self locates by detecting feature points in the image. However, it is difficult to extract effective feature points in the lack of texture environment, which is easy to cause tracking loss. In order to improve the robustness of slam in a thousand texture environment, some researchers put forward the slam method which combines feature points and feature lines. However, the extraction and matching of feature lines in these methods will consume a lot of computing resources. The paper proposes a direct tracking algorithm for line segment features, LSD algorithm is first used to extract the initial feature line segment in the key frame; in the ordinary frame, the direct tracking of line segment is realized by sampling, tracking and fitting steps. The algorithm is applied to the visual odometer, and experiments are carried out on the data sets of EuRoC and KITTI. The results show that the algorithm can improve the running speed of the visual odometer to 1.5 times compared with other algorithms, while maintaining similar trajectory accuracy.