Research On 3D Vision Positioning Technology Of Indoor Robot

The localization of a mobile robot is the essential element of its navigation task.Its goal is to ascertain the robot’s precise position in the global map,thereby allowing for precise positioning tracking.The pose search in the entire scene space,necessitating a great deal of computation,is a necessity for the global self-positioning process.However,the potential for scene ambiguity and dynamic shifts in local scenes during the positioning process can lead to the breakdown of prior data association,thereby diminishing the precision of positioning and even the effectiveness of global positioning.The two-dimensional image matching positioning technique can swiftly ascertain the robot’s global attitude,yet its accuracy is not particularly high.Generally,this is the initial step of visual positioning;however,the three-dimensional point cloud positioning technique can completely exploit the environmental structure data to acquire a very precise robot attitude,which is usually employed for more exact positioning.This paper delves into the 3D visual positioning technology of indoor robots,utilizing two distinct methods to enhance accuracy and efficiency.The research content of this paper is as follows:Firstly,in the process of stereo matching in robot vision,the stereo matching technology based on the visual word bag model in two-dimensional vision is introduced.Then,the stereo matching algorithm in 3D vision is introduced.In this stereo matching algorithm,the traditional SURF algorithm,FAST algorithm and SURF based polar line matching algorithm are described respectively.Experimental comparison shows that the pole-line matching algorithm based on SURF has higher performance.Combined with the Pn P algorithm to estimate the position of the current robot,the attitude position of the current robot can be better obtained.Secondly,in the stage of three-dimensional point cloud registration,the traditional ICP algorithm is described first.Its shortcoming is that the point sets obtained from different perspectives only partially overlap each other,so it is easy to obtain the local optimal case.However,GICP algorithm relies on nearest neighbor search,which makes it difficult to process a large number of point cloud data in real time in a computer with limited resources.This paper proposes an improved GICP algorithm to address the deficiencies of the two preceding ones,which is capable of point cloud stitching by extracting the corresponding point pairs of image features and mapping the image-point cloud relationship.Beginning with SURF’s pole-line matching algorithm,image features were matched,followed by the RANSAC algorithm to calculate the initial transformation matrix.Finally,the ICP algorithm was utilized to determine the precise transformation parameters-this is the fundamental procedure.On the one hand,the improved GICP algorithm can provide relatively accurate feature point pairs and improve the solving accuracy;on the other hand,the extraction method of feature matching point pairs is simpler,and the calculation amount is reduced to a large extent.Experimental comparison shows that the improved GICP algorithm has higher efficiency and better robustness.Finally,propose an improved Monte Carlo localization algorithm.In the improved Monte Carlo localization algorithm,this paper first introduces the traditional Monte Carlo localization algorithm,summarizes its process and analyzes its shortcomings.Then,an adaptive observation model is emphasized,which can be used for image matching.Finally,the resampling process of Monte Carlo localization algorithm is introduced,and the principle of AMCL algorithm in solving particle degradation problem is explained in detail.At the same time,feasible solutions to robot abduction and localization failure problems are provided.The experimental results show that the improved algorithm proposed in this paper is better in terms of speed,the adaptive observation model gives more reliable weight to particles and improves the convergence efficiency of particle sets,and the coupled response mechanism of robot abduction problem can quickly recover the effective positioning process.