Simultaneous Localization And Mapping Of Mobile Robots In Indoor Environment Based On Kinect

Simultaneous localization and mapping technology has been researched for nearly 30 years,and it has always been a research hotspot of robots and computer vision.Precise localization and mapping are prerequisites for mobile robots to autonomously locate and navigate in an unknown environment and complete people's assigned tasks.This paper uses the Qbot2 robot as a platform to study the related algorithms of SLAM(simultaneous localization and mapping,SLAM for short)in indoor environment using Kinect vision sensor.Mainly includes the following aspects:1.According to the structure of Kinect vision sensor and the accuracy of acquiring data,it is calibrated and registered by the checkerboard method.Then the obtained depth map is denoised and analyzed with different filtering algorithms.2.Aiming at the problem that ORB feature matching has low precision,this paper proposes a new culling algorithm for mismatched pair.In the algorithm,for the matched feature point pairs,firstly,the appropriate hamming threshold is selected by the experimental design to perform coarse culling,then the difference matrix is constructed by using the values of the surrounding pixel points to generate the description vector,and finally the correlation of the space vector is utilized.Sexual culling,and finally proved the effectiveness of the algorithm through experiments3.For the problem of cumulative error in the process of image frame registration,this paper introduces the loopback detection and graph optimization theory,realizes the loopback detection algorithm based on inter-frame registration and image appearance information,and utilizes the position of the robot.The pose relationship between the pose and the adjacent frames and the loopback frame constructs the pose diagram of the robot,and finally solves it using the g2 o general graph optimization framework.Experiments show that after the loop detection and graph optimization,the cumulative error of the robot pose is significantly reduced.4.With Qbot2 mobile robot as the platform and Kinect as the visual sensor,the effectiveness of the mismatch culling algorithm proposed in this paper is verified by experiments,and the culling accuracy is higher than the RANSAC algorithm.Then the SLAM experiment is used to compare the spatial point cloud map construction effect of the robot with or without loop detection and graph optimization.The validity and accuracy of the algorithm are analyzed according to the effect of the mapping.