Research On Map Creation Method Based On Binocular Stereo Vision For Mobile Robot

With the wide application of mobile robots in fields such as education and entertainment,environmental detection,medical services,and national defense security,new demands are placed on the intelligentization and autonomy of mobile robots.Simultaneous Localization and Mapping(SLAM)is the most important challenge for mobile robots to achieve their autonomy.The SLAM contains two problems of positioning and mapping.The existing SLAM systems mostly study the positioning problems and do not pay much attention to the construction of the map module.The established sparse feature point map mainly serves the positioning problem.In specific applications,the map is not only used to assist in positioning,but also has other requirements,such as: robot path planning(navigation),obstacle avoidance,and so on.Therefore,there is an urgent need for how to construct an environment-friendly map efficiently.The research in this paper is to use only CPU to realize the map construction of indoor and outdoor environments.This research has important practical value for the application of mobile robots.The study is as follows:In the cost aggregation method based on tree structure,the weight support region is selected by color information,and it is easy to produce mismatched in the image boundary area.A variable weight cost aggregation stereo matching algorithm based on horizontal tree structure is proposed.The initial disparity value is obtained by the horizontal tree cost aggregation,the horizontal tree is reconstructed with the initial disparity value and the color information,and the disparity map is obtained on the updated tree structure.In the disparity refinement stage,an improved non-local disparity optimization algorithm is proposed by introduce the pixel points that do not satisfy the left-right consistency constraint into the cost volume,which improves the correct rate of the final disparity map.Performance evaluations on all 31 Middlebury stereo pairs and the experimental results demonstrate that the proposed algorithm achieves an average error rate of 6.96% in non-occluded areas without disparity refinement and cost aggregation cost 1.52 s.In the field of indoor mobile robots synchronous positioning and 3D dense map creation system,most of the existing methods rely on GPU parallel computing to reconstruct dense three-dimensional map.However,in many real-world embedded systems,may not include GPU or GPU for other purposes.Therefore,in order to solve the problem of how to only use the CPU to recover the three-dimensional dense map of the environment,A binocular 3D dense construction algorithm based on ORB-SLAM2 is proposed.The algorithm is divided into four threads: trace thread,local map thread,closed loop detection thread and dense mapping thread.Dense mapping thread based on the known keyframe pose,stereo matching of binocular images to obtain the initial depth map,inverse depth fusion to obtain an optimized depth map.Stitching the depth images on different keyframes to create a dense three-dimensional map of the environment.Experiments with KITTI datasets show that the average running time of the algorithm in this paper is about 1 second,and can effectively establish a three-dimensional dense map that is consistent with the real scene.In order to verify the practicality of the dense map created in this paper for the autonomous navigation and obstacle avoidance of mobile robots,the matching accuracy of stereo matching and the accuracy of the map are analyzed.In the stereo matching error analysis experiment,the performance of the matching algorithm is evaluated for different experimental scenarios.The experimental results show that the variable weighted cost aggregation algorithm has higher matching accuracy for the occluded and tilted plane regions of the image.Then the accuracy of the map is evaluated in terms of the density of the map,the operating speed of the map creation,and the accuracy of the trajectory of the camera.The experimental results show that the accuracy of the algorithm map satisfies the navigation requirements of the robot.