Dynamic scene SLAM algorithm based on semantic information and joint constraints of optical flow and geometry

Abstract

Traditional simultaneous localization and mapping (SALM) algorithms are based on static environments. If there are dynamic objects in the environment, it will cause inaccurate positioning or problems that cannot be located. In order to solve this problem, the method of SegNet lightweight neural network and sparse optical flow combined with multi-view geometry is proposed to eliminate dynamic feature points. Firstly, the SegNet network is used to obtain the mask of potential moving objects. Secondly, sparse optical flow and geometric methods detect dynamic feature points. Finally, the dynamic feature points detected by semantics, optical flow, and geometric methods are combined to reject the feature points. This method can improve the positioning accuracy of the SLAM system in a dynamic environment.