Research On SLAM Algorithm Based On Binocular Vision

Computer vision is one of the main research directions in the computer field.Among them,robot vision is an important part of computer vision.The core problem facing is the realization of the Simultaneous Localization And Mapping(SLAM)of robot.The current binocular visual SLAM algorithm was mostly for the construction of sparse maps,and the constructed maps are difficult to fully reflect the surrounding environmental information.Therefore,this thesis improves the binocular vision SLAM algorithm and constructs grid maps in unknown environments.This thesis mainly completed the following work:(1)This thesis analyzed the overall framework of visual SLAM.Through research on visual odometer of visual SLAM and Back End,an improved keyframe extraction method and an improved graph optimization algorithm were proposed.(2)This thesis studied the camera imaging model and the binocular camera imaging model,calibrated the binocular camera according to Zhang's calibration principle,and proposed an improved binocular stereo matching method.(3)This thesis researched on the ORB-SLAM2 algorithm.In order to construct a dense grid map,the map construction process and Loop Closing function were improved.The innovations of this thesis are as follows:(1)An improved keyframe extraction method and an improved graph optimization algorithm were proposed.Keyframes with richer environmental information were obtained,the number of redundant keyframes was reduced,and the speed of back-end optimization was improved.(2)An improved binocular stereo matching method was used to increase the calculation speed and the matching accuracy.(3)The map construction process of the ORB-SLAM2 algorithm was improved,and the grid map construction was completed based on the binocular camera.At the same time,in the loop detection,the text dictionary was expressed in a binary format,which improved the initialization speed of the system and reduced resource consumption.Based on theoretical research,this thesis built a complete visual SLAM system and performed experimental analysis based on ORB-SLAM2 using an improved keyframe extraction method,an improved binocular stereo matching method,and an improved map construction process.The results showed that the improved algorithm in this thesis,got relatively reliable map data,and provided effective map feedback.