Design And Implementation Of Automatic Carrier Vehicle Control System Based On Visual SLAM

In recent years,with the rapid development of computer information technology,the technology of autonomous mobile robots has also made great progress and improvement.In this process,SLAM technology is one of the most common and effective means to help robots realize the environment information perception.The purpose of this project is to study the application of visual SLAM technology in automatic carrier vehicle,and to lay a foundation for subsequent research and application.This paper combines the difficulties of project and the characteristics of automatic carrier vehicle using RGB-D camera as the main sensor,ROS system as the operating system of the vehicle and SLAM system with feature point method for positioning and mapping.This paper also makes corresponding adjustments and improvements for this application,mainly in the following four aspects:(1)The feature extraction part of the car has been adjusted and improved.The ORB algorithm doesn't solve the scale invariance effectively and extract key points unevenly,the improved scale space construction and double threshold gridding are applied to the image.The former makes ORB algorithm have a degree of scale invariance,and reduces the number of image pyramids required.Double thresholds are used to detect the critical points in the image to ensure that enough key points can be extracted in different regions.Then,the gridded image is filtered to make the key points extracted more evenly.(2)The mapping function of the vehicle has been adjusted and improved,and the appropriate navigation map has been selected and built.The original ORB-SLAM system provides a sparse point cloud map,which can't be used to vehicles directly for obstacle avoidance and routing.This paper uses point cloud montage to construct a semi-dense map of the external environment.The semi-dense map can provide more effective environmental information.However,point clouds still can't be directly used for vehicle.This paper uses octomap method to convert point clouds into a three-dimensional occupancy grid map,and then uses vertical projection to build a two-dimensional occupancy grid map.(3)The design of automatic carrier vehicle based on embedded system is carried out.The circuit principle design of the data processing end,the motion controlling end and the part of motor drive is finished.The motion analysis of the Mecanum wheel used is carried out,which leads the motion controlling end to finish multi-direction movement by motor control,and the incremental PID is used to achieve the closed-loop control of the motor.(4)The image feature matching,mapping and trajectory and path planning of the vehicle are experimented.The experiments test the function of semi-dense mapping and get track error with evo tool by data set testing and actual site testing.The experimental results show that the maximum absolute error of the two-dimensional occupancy grid map built in this paper is 0.12m and the relative error is 7.1%.The average absolute trajectory error is 0.072m and the relative position error is 0.013m.The point-to-point path planning functions are verified on the two-dimensional occupancy grid map with the A*algorithm.