Accurate Line Reconstruction for Point and Line-Based Stereo Visual Odometry

The point feature is widely used in feature-based visual simultaneous localization and mapping (V-SLAM) or visual odometry (VO) systems, whereas the line feature is rarely explored even though it is more robust and contains more structural information than the point featue. In this paper, we propose a method of 3-D line reconstruction for stereo visual odometry with point and line features. By modeling the uncertainty of a line feature with its structural properties, the proposed line reconstruction method could refine the spatial positions of highly-uncertain lines from different stereo camera frames. The novel refinement process makes it possible to leverage a coarser but faster line detecting algorithm without loss of accuracy for the system. In the back-end optimization, the proposed method parameterizes a spatial line in a compact and decoupled way and applys a unified cost function for both point and line measurements. Extensive experiments have been conducted on different datasets and the results demonstrate that the proposed method can reconstruct a more accurate map of environments and achieve a higher speed as well as an accurate camera trajectory.