Reserch On Indoor Localization Of Vision SLAM Based On RGB-D

With the development of today's society and the continuous advancement of science and technology,the service capabilities of mobile robots working in indoor environments are becoming higher and higher.The research on indoor mobile robots must first solve the problem of localiziton,but GPS signals cannot be obtained in indoor environments.The advent of SLAM(Simultaneous Location and Mapping)provides a new way to achieve accurate Location in indoor environments.SLAM technology enables the robot to build maps and locate automatically in an unknown environment.Laser SLAM and visual SLAM are the main directions of SLAM technology research.Compared with laser sensors,it is the high price and application conditions are more demanding,but vision sensors is low price,and the application range is wide.Vision sensors have become a hot research topic in visual SLAM.The RGB-D sensor can simultaneously acquire color images and depth images of the surrounding environment,so it is widely used in visual SLAM.This paper studies the indoor positioning algorithm of visual SLAM based on RGB-D data,and uses the Kalman fusion to fuse the optical flow tracking method and the feature point matching method.This paper studies the principle of visual SLAM and different feature point extraction and matching algorithms.Experimental analysis and comparison of several mainstream feature point matching algorithms are carried out,the faster ORB algorithm is selected for feature point extraction,and the brute force matching method is used for feature matching.In order to reduce the impact of mismatched points on camera motion estimation,this paper proposes an optimization algorithm to coarsely remove feature point pairs,and then combines RANSAC(Random Sample Consecouse)algorithm to eliminate mismatched points.However,the real-time performance of visual SLAM based on feature point matching is insufficient,which affects the camera motion estimation.In order to improve the speed of the algorithm,this paper studies the faster speed of optical flow tracking SLAM.The traditional LK optical flow tracking method and the improved LK optical flow method based on the image pyramid are studied.The difference between the two algorithm tracking feature points is experimentally compared,and the improved LK optical flow tracking method is selected for the experiment.At the same time,it is proposed to use the RANSAC algorithm to eliminate redundant points and reduce the calculation of camera motion estimation.The optical flow tracking method can guarantee real-time performance,but there are problems of error accumulation and loss of feature points,which cause a large deviation in camera motion estimation.In order to solve the problems of poor accuracy and error accumulation in optical flow tracking method,Kalman fusion of feature point matching method and optical flow tracking method is proposed to reduce the error of optical flow method and increase the speed of algorithm.By carrying an experimental platform,experiments were conducted based on data sets and real scenes,which verified the real-time and accuracy of algorithm positioning.Experimental results show that: the algorithm in this paper can overcome the shortcomings of optical flow tracking method,such as insufficient accuracy and error accumulation,and at the same time can increase the speed of feature point matching method,which can provide more accurate positioning information for SLAM.