Research On Visual SLAM Of Mobile Robot Based On RGB-D Camera

With the rapid development of artificial intelligence in recent years,simultaneous Localization and Mapping(SLAM),which solves the problem of robot environment perception,has a wide range of applications in areas such as autonomous driving,robot navigation and augmented reality.However,as one of the key technologies of autonomous localization and navigation,visual SLAM still faces many challenges,such as the images acquired by the camera are noisy,single environmental features are difficult to handle,and the system cannot run in real time.Therefore,after deeply studying the framework of classic visual SLAM,this paper uses RGB-D camera as a sensor to design a visual SLAM algorithm that can run on the lunar surface environment with single scene features and harsh lighting conditions.The main work and innovations of this paper are as follows:(1)This paper analyzes the basic theory of classic visual SLAM algorithm and the mathematical model of RGB-D camera in detail,and then puts forward the evaluation standard of visual SLAM algorithm.(2)After studying the difficulties and defects of the traditional ORB feature extraction algorithm in extracting environmental features,the ORB feature points were improved from the two aspects of FAST key points and BRIEF descriptors,and a grid uniform ORB feature extraction algorithm suitable for the lunar surface environment is proposed.In this paper,the performance test and comparison of SIFT,SURF,ORB and grid uniform ORB algorithms are respectively carried out in the real lunar surface pictures.Finally,it is verified that the improved ORB features designed in this paper can meet the requirements of actual working conditions.(3)After comprehensively comparing the performance of BFM,FLANN and RANSAC feature matching algorithms,a fast RANSAC algorithm based on prescreening is proposed in this paper for the problem of slow calculation and false matching of feature matching algorithms.The improved RANSAC algorithm eliminates the matching with larger error by setting an appropriate Hamming distance,and further filters for mismatches through secondary reverse matching to obtain the optimal matching point set.(4)At the back end of the visual SLAM,this paper uses the BA algorithm for local adjustment of pose,and the global optimization of pose and map through the graph optimization algorithm after loopback detection.Faced with challenges such as single environmental characteristics and insufficient word bag models,a loop detection algorithm based on the relative position relationship of feature points on key frames is proposed.In terms of map construction,this paper selects a space-occupying Octomap that is convenient for navigation and takes up little memory as the the map construction scheme of visual SLAM.(5)Experimental test: First of all,this paper verifies the effectiveness of modules such as visual odometer,back-end optimization and loopback detection.Secondly,the comprehensive performance of the visual SLAM algorithm and ORB-SLAM2 algorithm in this paper is quantitatively analyzed in the TUM dataset.The results show that the localization accuracy of this algorithm is improved by 5.7% on average compared with ORB-SLAM2.Thirdly,the overall verification of the visual algorithm in this paper was conducted in indoor small-scale and outdoor large-scale environments,experimental results show that the algorithm in this paper can achieve accurate localization and the construction of sparse environment map.Finally,in order to pre-study the motion control technology of the lunar teleoperation robot,this paper uses the Gazebo simulation platform to create a high-fidelity lunar simulation environment,load the robot model designed in this paper,and complete the visual SLAM algorithm test.The experimental results show the applicability and robustness of the visual SLAM algorithm designed in this paper.