Study Of Location Algorithm Of Indoor Mobile Robot Based On RGBD

The SLAM(Simultaneous Localization and Mapping)problem of mobile robots is a part of the core problems in the navigation of mobile robots,and it is the premise of the task of robot path planning.In recent years,vision-based SLAM has gradually become the mainstream in SLAM research.In this paper,Kinect is used as a visual sensor to study the essential algorithms for positioning of indoor mobile robots based on RGBD.Firstly,the commonly used feature extraction algorithm is compared and analyzed.ORB(Oriented FAST and Rotated BRIEF)is selected as the image feature,and an improved ORB algorithm based on region segmentation is used to extract features.Compared to traditional algorithms,the feature points extracted by this algorithm are more uniform.Secondly,in the feature matching,Brute-Force algorithm is utilized for preliminary feature matching,and a fast average distance method is proposed,which combines RANSAC(Random Sample Consensus)algorithm to remove false matching.This scheme is more adaptable and more efficient.Thirdly,in the keyframe selection strategy,this paper proposes a keyframe selection strategy for the optimal candidate frame through the study of the keyframe selection method for traditional motion estimation.First,the suspect keyframe is chosen as the candidate frame,and evaluate the optimal frame as a keyframe from several aspects.This method reduces the loss probability of image tracking and improves the robustness of the entire system.Finally,in the loop detection,a simple and effective loop detection strategy is proposed,and proposed a model for eliminating the mistaken loop closure.And optimize the pose graph by using the graph optimization method which can reduce the accumulated error of the robot pose.Experimental results of the proposed scheme based on the public benchmark dataset show that the positioning accuracy of this system can reach within 5cm,and the processing speed can reach 25 fps.Results of experiment under indoor real scene also show that the proposed scheme can estimate the trajectory of the robot accurately and build a 3D map of the indoor scene.