Research On SLAM Technology Based On Multi Feature Fusion And Graph Optimization Framework

Simultaneous Localization and Mapping(SLAM)is a hot research topic in the field of robotics,and it is also a key technology with very considerable application scenarios.In recent years,the visual SLAM scheme based on feature points has become more mature.However,the performance of this algorithm will decrease sharply when the feature points are scarce or unevenly distributed.To this end,this paper proposes a multi-feature fusion,combining points and line segments.The visual SLAM algorithm based on the filter method has the problems of high computational complexity,large environmental storage space load,linearization error and update efficiency.Therefore,this paper proposes to use graph-based optimization to solve SLAM.In a low-texture environment,it is often difficult to find enough reliable feature points,which leads to a decrease in the accuracy of the SLAM system or even a complete failure.However,in the low-texture environment,there are often rich linear state plane elements,so line segments can be extracted from these plane elements and combined with points.These two features will play a complementary role in the SLAM system,and then let the SLAM system More accurate,more robust,and more stable.In addition,an environment map composed of 3D points and line segments will have more structural information than a map composed of only points.In this paper,point features and line features are extracted from RGB-D images,and these two features are combined.Based on the graph optimization method,an RGB-D SLAM method that fuses point and line features is proposed.The main research work of this paper includes:(1)The principles related to visual SLAM,such as the principle of each spatial coordinate system and the conversion relationship between them,the principle of camera imaging,the camera calibration algorithm,the RGB-D camera model,and the distortion model,are studied.(2)Based on the method of parameterizing the spatial straight line by the end points of the line segment,the projection error model of the fusion point and the end point of the line segment is studied.In addition,for the case where there is one more feature descriptor,a specific visual dictionary and database for LBD descriptors are built based on Bo W technology,and how to use the advantages of point features and line features is studied.(3)Based on the graph optimization theory,an RGB-D SLAM algorithm is designed to fuse point features and line features.In the SLAM process,point features are extracted and described by the ORB algorithm,and line features are extracted using the LSD algorithm and described by the LBD algorithm.The descriptors for both image features are stored in binary form.