Research On Visual SLAM Algorithm Based On Dynamic Object Detection And IMU

Robotics is becoming more sophisticated as artificial intelligence advances.Mobile robots are increasingly being used in different development and life scenarios.Although the visual simultaneous localization and mapping(SLAM)algorithm is the core technology of mobile robots,most visual SLAM algorithms still have flaws.As a consequence,this study explores the SLAM approach based on IMU and stereo vision fusion,which has both theoretical and practical significance.The following is the thesis' core work:First,since most visual SLAM methods assume that the environment is static,dynamic objects in the scene can result in a lot of incorrect data associations during localization and mapping.To address this problem,this thesis proposes a visual SLAM method based on instance segmentation,which is built on the basis of ORB-SLAM2 with the addition of dynamic object detection,uses an instance segmentation model to segment the input image and identify dynamic objects,and generates a mask for masking dynamic feature points in the image to reduce the dynamic objects' impact on the pose estimation of the SLAM system impact.And,in high-dynamic scenes,the improved method outperforms the ORBSLAM2 algorithm,according to experimental results on the TUM dataset.Subsequently,this thesis integrates IMU information on the above SLAM system and proposes a pose estimation method with IMU and stereo vision fusion to address the deficiency that visual SLAM would have camera tracking failure in the case of rapid camera rotation,etc.The pose estimation process is modeled as a control system in this system.The vision and IMU information processing are based on feedback loops,which are the gradient reduction feedback loop and the bias estimation feedback loop,respectively,to achieve the stability of the system.And the experimental validation and evaluation are carried out on the Eu Roc MAV visual inertial dataset.Compared with the ORB-SLAM2 method,the IMU and stereo vision fusion method can obtain higher absolute trajectory accuracy,which effectively improves the robustness of the system.Finally,the vision-IMU SLAM system is designed,complete with IMU and vision information fusion.And,to test the localization and map building in static and dynamic scenes indoors,this thesis' method is implemented on a designed mobile robot.By comparing the localization-effects with ORB-SLAM2 algorithm,the results show that,as compared to ORB-SLAM2,the method in this thesis decreases the impact of dynamic objects and effectively increases the accuracy in dynamic settings,proving the method's feasibility.