Research On SLAM For Wheeled Robots Based On Depth Camera And IMU Fusion

Wheeled robots need to complete designated tasks in complex environments,which places higher demands on the autonomy of robot navigation.Therefore,studying SLAM technology is of great significance for the autonomous navigation of wheeled robots.Visual SLAM acquires environmental data using cameras as sensors.However,cameras cannot work when the robot is moving quickly or when the environment has low texture,and IMU sensors can compensate for the shortcomings of cameras.In addition,when there are dynamic objects in the environment where the robot is located,dynamic points in the environment will reduce the accuracy of pose estimation.To address the above two problems,the following research has been conducted:Firstly,this paper deeply investigates the odometer based on depth cameras and derives the IMU pre-integration process.As the noisy depth image data can affect the pose estimation,this paper proposes a pose estimation method based on the combination of EPn P and ICP,which improves the accuracy of the pose estimation.In addition,to address the harm of dynamic point pairs on pose estimation,this paper proposes a dynamic point removal algorithm for visual odometer.By filtering dynamic points twice using IMU pre-integration and epipolar constraints,the interference of dynamic objects on pose estimation is reduced,and the accuracy of pose estimation in dynamic environments is effectively improved.Secondly,this paper implements the tightly-coupled optimization of depth camera and IMU.After in-depth research on the visual reprojection error model and the IMU preintegration residual model,a tightly-coupled optimization model for pose optimization and local optimization is proposed.In the pose optimization model,prior information of the state variables at the previous moment is considered,and two methods are designed to optimize the pose according to the validity of prior information.In the local optimization,a sliding window is used to control the scale of the state variables to be optimized.A local optimization method is designed by combining visual constraints and IMU constraints,and optimized in a bundle adjustment(BA)manner,which improves the accuracy of localization and mapping.In the global optimization,pose graph optimization is performed first,followed by global optimization,which accelerates the optimization process.To address the problem of drift caused by error accumulation,the paper studies the calculation methods of bag-of-words model and image similarity,and corrects the global pose through loop closure detection,further improving the accuracy of localization and mapping.Finally,the proposed SLAM algorithm in this paper is tested.First,the depth camera and IMU are jointly calibrated.Then,the proposed dynamic point detection algorithm is tested,and the experimental results prove that the algorithm can effectively distinguish between dynamic and static points.To further verify the performance of the SLAM algorithm proposed in this paper,experiments are conducted in different scenarios in static and dynamic environments,with ORB-SLAM2 as a benchmark.The experimental results in static environments show that even in scenes with fast robot motion and simple textures,the proposed SLAM algorithm can still maintain accurate pose estimation.The experimental results in dynamic environments show that the proposed SLAM algorithm can reduce the interference of dynamic targets and achieve higher localization accuracy than ORB-SLAM2.