Research And Implementation Of Indoor Mobile Robot SLAM Based On Binocular Vision

In recent years,with the development of mobile robot technology and image processing technology,SLAM(Simultaneous Localization and Mapping)technology applied to mobile robots has gradually been favored by scholars at home and abroad.As the most similar sensor system to human eyes,binocular vision system can directly obtain the depth information of the environment.At the same time,it has the advantages of low cost and abundant information.Therefore,binocular SLAM has become a research hotspot nowadays.In order to solve the error problem of a single sensor,Multi-sensor fusion will become the research trend of SLAM in the future.Therefore,the research of binocular sensor and IMU fusion is of great significance.This paper mainly studies SLAM of mobile robot based on binocular vision in indoor environment.There are two aspects: on the one hand,it simplifies the computation of SLAM algorithm;on the other hand,it optimizes the pose of the robot by the fusion of binocular sensors and IMU.The research contents are as follows:Firstly,the commonly used SLAM schemes are analyzed.The advantages and disadvantages of each scheme are compared,on this basis,binocular vision SLAM is chosen as the research direction,ORB-SLAM2 is chosen as the basis of the algorithm.Secondly,the algorithm structure of ORB-SLAM2 is studied.It includes coordinate system relations,feature points extraction and matching,robot pose calculation and optimization,etc.It provides a theoretical basis for subsequent data fusion.Then the image distortion model is analyzed,the binocular camera was calibrated by Zhang Zhengyou calibration method.Aiming at the problem of large computation of ORB-SLAM2 algorithm,an optimization scheme is proposed to pre-position the distortion processing process into the camera.Before optimization,SLAM algorithm needs to correct image distortion first,it takes part of the time,Increased CPU load,by transferring the distortion processing flow,to a certain extent,it reduces the computational complexity of the algorithm.,it relieves the pressure of CPU operation.Finally,the optimized mathematical model is deduced,and the optimization scheme is verified by using EuRoc data set and recorded data package.The feasibility of the optimization scheme is proved by the comparison of operation time.Pure binocular SLAM has poor image reliability,there is a phenomenon of displacement.In order to solve this problem,this paper proposes a scheme based on loosely coupled IMU(Inertial Measurement Unit)and visual odometer fusion.Firstly,the "camera + IMU" system is jointly calibrated,The corresponding relationshipbetween binocular coordinate system and IMU coordinate system is obtained.Based on Extended Kalman Filter(EKF),the fusion model of binocular visual odometer and IMU is deduced.The pose of robot is modified by integrating IMU data with visual odometer.The experimental results show that the fusion pose is closer to the real value.