Real-time RGB-D Slam With Points And Lines

Simultaneous Localization and Mapping(SLAM)is regarded as one of the core technology for automatic robots and self-driving cars,which has proven its value and gain much attention recent years.However,problems such as changeable illumination and low-textured environment make SLAM algorithms unavailable,especially for feature-based visual SLAM.The quality and the number of point feature drop in these situations.Therefore,a real-time visual SLAM with points and lines is proposed in this paper to address these problems.Specifically,our SLAM is built upon ORB-SLAM,and line features are added combining with point features to increase the accuracy and robustness of the SLAM algorithm.The main contributions of this paper include:(1)Line-segment-detection(LSD)algorithm and the corresponding line features matching,optimization and elimination algorithms are added.Extraction of ORB feature point is modified,such as its FAST corner is replaced by AGAST corner,and its implementation of rotation invariance is modified to reduce the influence of illumination changing.(2)A method of measuring the stability of point and line features is proposed to reflect the quality of point and line features.The value of the information matrix in graph optimization is adjusted according to the stability and the information of the point and line features.(3)The visual SLAM algorithm with points and lines is implemented completely,including the acquisition,calibration and filtering of stereo imaging data,the extraction and matching of point features and line features,localization,local mapping,loop closing,graph optimization algorithms,and so on.The experimental results show that compared with the original ORB-SLAM algorithm,the proposed algorithm has better accuracy and robustness,and can comply with the requirements of real-time operation.