Research On Monocular Vision-Based AGV Guiding Technology

Automated Guided Vehicle(AGV),as an important tool with wide application in manufacturing,warehousing and other scenarios,undertakes the task of material handling and transportation in intelligent logistics.With the further development of intelligent logistics technology,the scope and field of application of AGVs has been expanding,and new demands have been made on the autonomy and intelligence of AGVs,which are highly dependent on the accuracy of positioning information.In general,AGV positioning accuracy is mainly affected by the accuracy of its own installed sensors,but the excessive pursuit of high-precision sensors,their expensive prices invariably reduce the competitiveness of AGV products.However,when the camera runs with the AGV at variable speed,the blurred image caused by the constant camera frame rate can easily lead to the loss of positioning information.In order to obtain continuous positional estimation,the paper proposes a loosely coupled localization scheme based on ArUco marker vision localization with Li DAR and Inertial Measurement Unit(IMU),and the main work of the paper is as follows.(1)For the problem that visual 2D code localization is easily affected by the environment and marker mis-detection,the ArUco marker-based indoor localization method is proposed.Firstly,according to the structural characteristics of the ArUco marker black boundary surrounding the internal encoding matrix and ID encoding constituting a dictionary,adaptive thresholding is used to realize image segmentation and extract quadrilateral contours.Secondly,by setting the perimeter threshold,the extracted ArUco contours are filtered for the first time to obtain a set of candidate contour collections.Each candidate contour bit in the set is extracted and decoded,and the contours that are correctly decoded and match the dictionary content are retained for the second screening.Then,we extract the sub-pixel coordinates of the four corner points of the black boundary of the ArUco marker,solve the camera external reference matrix based on the multi-point perspective positioning(Perspective-n-Point,Pn P)algorithm,and construct a local static map by setting the reference coordinate system and minimizing the reprojection error principle to realize the unification of the ArUco marker coordinate system.Finally,based on the constructed static map,indoor positioning of AGV based on monocular vision is realized.(2)To address the problem that the cumulative error of the traditional odometry estimation method is difficult to eliminate,the Range Flow Based 2D Odometry(RF2O)algorithm is used instead of the traditional odometry for the planar motion estimation of AGVs.To reduce the errors of RF2 O algorithm due to mechanical interference and environmental static assumptions of range flow equation,we realize the improvement of positioning accuracy by establishing AGV motion model and observation model and fusing Li DAR and IMU with Extended Kalman Filter(EKF).(3)To address the problem that the LIDAR indoor global matching localization method is prone to localization errors and insufficient reliability in environments with few or similar feature points,the localization results based on EKF fused RF2 O and IMU are corrected with the positional information provided by ArUco markers and fed into the Adaptive Monte Carlo Localization(AMCL)localization system for particle state update to realize the loosely coupled localization method of ArUco marker with LIDAR and IMU.(4)Design and implementation of AGV software module based on ROS system.The software system is equipped on the AGV hardware platform,and the positioning algorithm proposed in the paper is verified through experiments.The experiments show that the loosely coupled localization scheme based on ArUco markers with Li DAR and IMU is practical and feasible,and the root-mean-square error is significantly reduced in the case of shorter distance operation.