Research On Indoor Visual Localization Algorithm Based On KINECT2.0 And ORB_SLAM2

Simultaneous localization and mapping(SLAM)is the key technology of mobile robot application.Vision sensor has the advantages of low price and abundant information,so the SLAM technology based on vision sensor is paid more attention.But for VSLAM,it is one of the difficulties in this field to deal with complex and various information to realize the real-time and accurate positioning of robot.This paper studies the indoor vision localization algorithm based on KINECT2.0 and ORB_SLAM2.On the basis of the basic theory of VSLAM,combined with the characteristics of ORB feature points,and making full use of the characteristic of RGB-D camera which can directly obtain the depth of the image,it realizes the accurate and better real-time performance of visual positioning.In order to effectively improve the efficiency of visual positioning,this paper proposes an improved PROSAC algorithm to process feature data,so as to avoid the blindness and low efficiency of RANSAC algorithm in the process of feature tracking.The PRO SAC algorithm first calculates the sample score,then arranges the samples in descending order according to the sample score,extracts the samples from the sample subset with higher quality,and solves the model parameters to obtain higher calculation efficiency.In addition,due to the large number of key frames and map points involved in the global BA optimization,which leads to much computation,this paper proposes an improved global BA optimization algorithm.On the one hand,the normalized cut algorithm Ncut is used to segment global BA for map points;on the other hand,the more efficient LDLT matrix decomposition algorithm is used to replace the original Cholesky matrix decomposition algorithm,so as to obtain more efficient global optimization.Finally,this paper applies the above improved algorithm to visual SLAM system based on KINECT2.0 and ORB_SLAM2,and validates and analyzes it in data set and real-time scene.The experimental results show that after adopting the improved feature mismatch elimination algorithm based on PROSAC algorithm,and the global BA optimization method based on the normalized partition Ncut algorithm and LDLT matrix decomposition algorithm,the optimized time efficiency is nearly one time higher than that of the traditional ORB_SLAM2 system,and the positioning accuracy is 0.001m,which guarantees the positioning accuracy.In this paper,the system not only ensures the positioning accuracy and stability,but also improves the real-time performance compared with ORB SLAM2 system.