Fast Stereo Visual Odometry with Point-line Features Using Improved EDLines Algorithm

Abstract

Traditional point feature-based visual odometry (VO) makes it difficult to find reliable point features to estimate the camera pose in low-texture environments, resulting in a significant decrease in the positioning accuracy and robustness of the system. To address the above issues, we integrate line features into VO based on point features to improve the performance of the system in low-texture scenes. Specifically, we adopt an adaptive line feature extraction strategy based on the richness of scene texture information to solve the problem of difficulty in extracting sufficient point features in low-texture scenes while ensuring the real-time performance of the system. Then, we propose a line segment merging algorithm to improve the EDLines algorithm (LM-EDLines), making the extracted line segments more complete and improving the quality of line features. To reduce the positioning error of the system when the camera turns or changes speed sharply, we propose a motion model selection strategy based on error analysis. Finally, the experimental findings on the KITTI and EuRoC datasets demonstrate that the suggested technique outperforms previous state-of-the-art systems in terms of overall performance.