Simultaneous Localization And Mapping Of Indoor Mobile Robots Based On Semi-direct Method

With the research and development of indoor mobile robots,sweeping robots and other applications in our daily lives are becoming more and more popular,and the accuracy requirements for robot positioning and mapping are getting higher and higher.How to make the positioning and mapping of mobile robots in indoor environments more accurate,with smaller errors,is a problem we have to face,and it is also particularly important to increase the robustness of the system in complex and extreme scenarios.This paper proposes a depth visual SLAM system based on the semi-direct method.The main research content are divided into the following parts:Firstly,the standard pinhole projection model of the depth camera used in this article is introduced,and the camera is calibrated in consideration of camera distortion factors.Then the three-dimensional Euclidean motion transformation model of rigid body is described,and Lie algebra is introduced to represent the pose.The mathematical motion model of the visual SLAM system in continuous time state is established,and the uncertainty of system variables is described by the method of probability.Secondly,the visual odometry methods based on feature point method and direct method are studied in detail,and their advantages and disadvantages are analyzed.The SVO algorithm is introduced.Based on its idea,this paper proposes a new depth visual odometry method based on the semi-direct method.This method inherits the characteristics of both feature and direct based methods,and uses the images collected by the direct tracking depth camera to minimize the weighted error function proposed.The function takes the photometric error and the depth error into account to solve the initial pose of the camera.Secondly,the pose is locally optimized by minimizing the reprojection error,and the local map is constructed to increase the constraints,so that the camera's pose estimation is more accurate.Thirdly,the performance of the system is further optimized on the basis of the designed semi-direct depth visual odometry.In view of the large trajectory error of the algorithm in the dynamic scene,an improved moving object segmentation method is added to the front end of the visual odometry to process the dynamic scene of the image frame,remove the influence of moving objects,and increase the robustness of the system in the dynamic scene.It also introduces the back-end optimization and loop closure detection links to build a complete visual SLAM system while reducing the accumulated drift error caused by long-term motion.The qualitative and quantitative analysis of experimental comparison verifies that the optimized SLAM system based on the semi-direct method has greatly improved performance compared to a pure visual odometer.Finally,the standard data set is used to compare with some current mainstream visual SLAM algorithms,the experimental results show that the algorithm proposed in this paper has higher precision and accuracy in trajectory tracking,and the constructed 3D dense point cloud map has better effect.Build a mobile robot platform to conduct actual experiments in complex and extreme places such as indoors and corridors.The algorithm in this paper can accurately estimate the motion trajectory of the robot,and build a good laboratory environment map to verify the robustness and feasibility of the algorithm.