Research On Localizing Indoor Wheeled Robots Based On Visual-Inertial Odometry

With the development of autonomous driving,industrial robot,unmanned aerial vehicle in recent years,visual Simultaneous Localization and Mapping(VSLAM)and Visual Odometry technology are attracting great attention from researchers.However,in the artificial environment where the texture is not rich and the illumination changes obviously,the SLAM/VO system based on point features will lead to large error in system state estimation due to its scarcity or difficult matching.In this paper,a large number of structural lines in the indoor environment are selected and added to the system to make up for the lack of point features.In addition,the inertial measurement unit(IMU)and visual camera are integrated into the VIO system to give full play to the advantages of IMU’s rapid response.Aiming at the complex and diverse indoor environment,this paper puts forward the multiply Manhattan world assumption to accurately model the indoor environment.By clustering the extracted line features,the structural lines are filtered and the vanishing points are obtained.Then the vanishing points are associated with the Manhattan world direction,so as to obtain the rotation transformation between the camera and the current environment.The VIO system proposed in this paper mainly includes two modules:measurement processing and sliding window optimization.Measurement processing mainly extracts and initializes the corresponding point and line features from the image,and then aligns them with the results obtained by IMU pre integration.The sliding window optimization part is responsible for updating the measurement constraints after initial alignment,and ensuring the dimension of state quantity through marginalization strategy,so as to meet the requirements of real-time performance of the system.Finally,the experiment results of the VIO system proposed in this paper on the euroc dataset are 10-15%higher than the current mainstream VIO system(VINs and PL-VINs).Three experiments in the real environment show that the proposed VIO system can deal with most complex indoor situations and has high positioning accuracy.