Research On Optimization Of SLAM Graph Based On Point Cloud Data

With the continuous development of science and technology,autonomous navigation of mobile robot has become one of the research hot-pots.SLAM(Simultaneous Localization and Mapping)is a kind of technology that robot estimates its position in real time and creates a map of the surrounding environment according to the data acquired by sensors in the unknown environment.The vision based SLAM technology has attracted the attention of researchers because of its easy to extract feature points,rich feature information,convenient operation and important application value.At present,the vision SLAM Based on Kinect sensor can directly obtain depth information,thus reducing the amount of calculation,and becomes the mainstream of research on SLAM technology.In the SLAM visual odometer,there are many problems in the feature matching process,such as large error accumulation,inaccurate matching and long iteration time.In view of this phenomenon,on the basis of widely used RANSAC algorithm,by adding the setting of previous judgment processing link,the point pairs that do not meet the previous judgment conditions can be discarded,so as to reduce the error accumulation in the process of eliminating mismatches.The ICP Iterative Method based on BA is implemented to reduce the time loss.In the back-end optimization of SLAM,aiming at the disadvantages of the nonlinear optimization methods in the current visual SLAM technology,such as complex process and slow optimization speed,based on the framework of the widely used BA nonlinear optimization method,the core descent strategy Levenberg-Marquardt Method is used to limit the initial trust region and the proposed approximate range for the nonlinear graph optimization of the back-end of slam,making up for BA deficiency of nonlinear optimization,in order to achieve the ultimate goal of the overall optimization of the system..By comparing the experimental settings and analyzing the simulation results of the experiment,it can be concluded that the optimized RANSAC algorithm,compared with the traditional algorithm,can obtain more pairs of points after the elimination of mismatches,higher number of correct matches,and effectively improve the timeliness of the whole process;the optimized L-M descent strategy can speed up the optimization speed and improve the effect of map building.