Research On SLAM Technology Based On Binocular Visual Inertial Odometer

Navigation technology plays an important role in the development of mobile robots,and the foundation of mobile robot navigation is to locate its own location under movement.At present,most locating technologies need to have prior information or arrange tracks or marks in advance,and that limit its application.Therefore,how to accurately and stably achieve its own location without prior information is an urgent problem to be solved.In response to the above requirements,this paper researches the visual SLAM technology that can meet the above requirements.Pure vision has a poor estimate of short-term fast motion,the IMU can respond to short-term fast motion estimation,making up for the lack of pure vision,and the visual sensor in turn corrects the IMU drift.Therefore,the final research object of this paper is the visual inertial SLAM technology which combines the information of visual and inertial sensors.The main contents and specific work of this paper are as follows:(1)The feature point extraction and correlation algorithms for the visual front end are compared and analyzed.In the feature point extraction: for the feature distribution of the extracted feature points in the ORB algorithm mentioned in the opencv library.A double threshold detection ORB algorithm based on grid segmentation is selected.In the feature point correlation: the feature point correlation effect between the multi-layer optical flow tracking method and the feature point matching method are compared.It is found that the feature point matching method is more robust and accurate in the scene where the camera rotation and translation are slightly larger,so feature matching method is selected as feature point matching method.(2)Aiming at the existence of moving objects or pedestrians during the operation of SLAM,a feature extraction method based on two-dimensional pixel motion compensation is proposed,which Effectively detect and remove feature points from dynamic objects or pedestrians and reduces the interference of moving objects or pedestrians on the system and improves the accuracy of the SLAM system.(3)About the fusion of visual odometer and inertial sensor,based on the pre-integration theory of IMU,the visual-inertial fusion odometer joint initialization and back-end tightly coupling nonlinear optimization are studied,and improving the key-frame generation algorithm in the system,the combination of IMU information integration and improved keyframe algorithm effectively improves the robustness of algorithm in the scene with rotation or a small amount of motion blur.(4)The positioning accuracy of the designed SLAM algorithm based on binocular visual-inertial odometer is verified.Firstly,the public dataset is used to compare and analyze the proposed algorithm and the stereo vision algorithm.The results show that proposed method has improve the accuracy and robustness of the algorithm.Then the camera is mounted on the mobile robot or handheld.In this way,the positioning effect of indoor and outdoor real scenes is verified.The experimental results show that the positioning accuracy of the algorithm in indoor and outdoor is within the acceptable range.