Fast RGBD-ICP with bionic vision depth perception model

Abstract

How to improve the real-time performance of 3D SLAM(Simultaneous Localization And Mapping) is a key issue of mobile robot. In this paper, a bionic vision depth perception model is researched aimed at real-time performance of RGBD SLAM, which takes the sensors's depth value as a parameter. As we all know, it is more clear as the distance closer and more obscure as the distance further of biological vision, scene over the depth of field can be negligible. According to that a gradient filter algorithm of point cloud based on sensors's depth value is researched, which can reduce the calculated cost of ICP(Iterative Closest Point) and improve RGBD SLAM efficiency to get high quality of 3D map. Three kinds of experiments based on bionic vision depth perception model are discussed. Compared it with fast random sampling algorithm, the experimental results show that the bionic vision depth perception model greatly improves the real-time performance of RGBD SLAM.