Research On Indoor RGB-D SLAM Algorithm Based On Fusion Of Point And Plane Features

In recent years,the visual SLAM technology,which has the characteristics of rich perception information,low cost,and convenient portability,has received extensive attention.In the indoor environment with weak texture and multi-structure,limited by the high dependence of point features,the multi-structured features in the indoor environment cannot be fully utilized.In view of the above problems,this paper proposes an RGB-D SLAM algorithm that fuses point and surface features.The motion trajectory of the camera is effectively estimated by the fusion of point and surface features,and a global consistent map is constructed at the same time.Thus,the positioning accuracy and mapping accuracy of the mobile robot in the indoor weak texture multi-structured environment are improved.The main work of this paper is as follows:1.An improved AGAST feature extraction algorithm is proposed.Firstly,a comparative analysis of several existing image feature point extraction algorithms is carried out,and it is verified that the AGAST feature point extraction algorithm is better than the other three feature point extraction algorithms in real-time performance.Secondly,in view of the problem that the feature points are not evenly distributed in the image,it is easy to cause excessive redundancy after feature point extraction,and the method of quadtree is used to improve the image feature point extraction algorithm,and finally the feature points are matched by the Hamming distance and RANSAC algorithm.2.Propose an indoor mobile robot positioning and mapping algorithm based on point and plane feature fusion.Use the integral graph to calculate point cloud normal vector to perform plane segmentation of the ordered point cloud,and then calculate its vertical or parallel pseudo-planar features by segmenting the plane profile features,and combine the AGAST feature points to construct a structural constraint factor map for point-surface feature fusion,and derive the Jacobian matrix which solved the state estimation.The key frame selection method is established based on the image similarity and time and space rules to ensure the difference and traceability of the information between the selected key frames.According to the g2 o optimization library,the optimization of the factor map is completed,and the local map and the global map are realized.Optimize update management and loopback,and generate a globally consistent 3D dense point cloud map and an octree map that is more conducive to mobile robot storage.3.Using the TUM data set to verify and analyze the point-plane fusion algorithm’s positioning and mapping capabilities in an indoor environment on the embedded development board NVIDIA Jetson Nano.The absolute trajectory error and the relative trajectory error are used as the basis for judging the performance of the SLAM system.Compared with the ORBSLAM2 algorithm that only relies on point features,the point-surface fusion algorithm has better trajectory estimation accuracy for mobile robots in indoor environments than ORBSLAM2 Algorithm,and in the ORB-SLAM2 tracking lost data set,the algorithm in this paper can still maintain good tracking stability and accuracy,and finally effectively construct a 3D dense point cloud map of the environment and an octree map that is easy to store and reuse.