A Graph-based Visual SLAM System Fusing Multiple Feature

To reduce the strong dependence of feature matching of point-based visual simultaneous location and mapping system, besides, improve the efficiency of feature matching of line-based V-SLAM, a novel visual SLAM system named Multiple Features-SLAM by fusing the point, line segment and plane features is proposed. Firstly, AT-FAST algorithm by adopting adaptive threshold is used to solve the problem that traditional FAST features hold feature block. I-LSD algorithm is used to extract line segments, and improve the matching accuracy by reducing the redundancy of line segments. Extracting candidate plane based on the above features and using MPR to express plane segments to reduce the amount of calculation. We structure the error function by fusing point features, line segments and plane segments under g2o framework and obtain the convergent results. The experiments on TUM show that our algorithm can improve the positioning accuracy by 37.8% compared with some open source frameworks such as ORB-SLAM. The practical experiments show that the accuracy of our algorithm reaches 6cm and has strong robustness.