Vision SLAM For Indoor Wheeled Robot

In recent years,as researchers have continued to deepen the research on visual SLAM(Simultaneous Localization and Mapping,SLAM),the technology has become more and more widely used in industrial and civil applications.Based on the visual SLAM framework,this paper explores the unknown environment through wheeled robots equipped with RGBD((RGB-Depth)cameras).On this basis,this paper focuses on the visual odometer of planar motion wheeled robots.First of all,this article has done a relevant research on the accuracy of the depth value of the RGBD camera,and evaluated the ranging performance of the depth camera through experiments.The experiment shows that the camera is more suitable for use in small scene mapping.In the visual odometer part,this paper uses the ORB algorithm to extract and match the feature points of the adjacent two frames.According to the motion characteristics of the wheeled robot,the contour constraint conditions and the straight line constraint conditions are proposed.Realize the screening of feature points.The experimental results show that the constraint condition can effectively filter out the mismatched point pairs that occur during the matching of the inter-frame images,and it can also filter out the correct color map matching but large errors in the depth value.Mismatched point pairs can provide a high-quality matching point pair for the subsequent calculation of the motion between frames of the visual odometer.Because the planar wheeled robot moves in a two-dimensional plane,this paper performs dimension reduction processing on the obtained matching point pairs,and combines the two-dimensional ICP algorithm to estimate the motion trajectory of the robot's camera.This paper uses the TUM data set to verify the proposed wheeled robot visual mileage calculation method,and compares it with the currently commonly used RANSAC algorithm and three-dimensional ICP algorithm.The experimental results show that the overall operation of the ICP algorithm after dimension reduction is processed.The speed has been greatly improved.The estimation of the camera motion trajectory obtained by the visual odometer has accumulated errors,and a more accurate camera pose and motion trajectory can be obtained through loop detection and back-end optimization.Finally,this article uses the optimized camera pose and the point cloud data of the RGBD camera to construct a point cloud map,and on this basis,further constructs an octree map.