Research On The Key Technology Of Intelligent Vision Archive Access Robot

In order to realize the unattended management of the archival repository and improve the intelligence of archival management,this paper launches the research on the vision technologyled archival access robot.Targeting at the key technologies of autonomous navigation in the archival repository and archival box finding,as well as the archival box deposit and fetch under the conventional placement conditions by robot,the content of researches are as follows:Firstly,a method of robot navigation and addressing inside the archival repository is designed by integrating laser SLAM technology and QR code navigation method.At start,we design the integration system of LIDAR and robot classis cart,and through this system,2D point cloud data is collected to demonstrate the map of the archival repository;by embedding the path planning algorithm,the robot can address the object archival shelf by itself.And then,for the defect that LIDAR cannot collect point-cloud data inside the archival shelf,we use visual QR code navigation method to solve the posture and position correction when the robot walks for the search of object archival box,realizing the full-coverage autonomous navigation of the robot in the archival repository.At last,based on the barcode management method,a method of scanning and recognizing barcodes during the movement of the robot is designed to enable the robot to complete the search of archival boxes simultaneously during autonomous navigation.Secondly,vision-based methods for precise positioning and measurement of archival boxes are investigated.The explorations of the methods are carried out from the industrial traditional vision field and popular artificial intelligence vision field nowadays,respectively.One is based on the special color feature of archival boxes.A convolutional method combining morphological ideas is designed for the pixel block feature extraction,and then a neural network is used to classify pixel blocks with different color feature.By the composition of the pixel blocks,regions where archival boxes are located in the image can be outlined precisely,which greatly improves the accuracy of positioning and measuring archival boxes by Blob analysis method.The other is based on the deep learning target detection method,by adding angle inference and GIo U layer into the original YOLOv3 network,the network now can detect dumped archival boxes with better evaluation of the overlapping and rotating detection bounding boxes.By removing the redundant overlapping bounding boxes with EM algorithm,the positioning and measurement of archival boxes in complex densely packed scenes can be realized well.Thirdly,the hardware and software architectures of the overall archival access robot are designed.Considering the vision technology applied and the characteristic of archival boxes,a robotic hand consists of vacuum suction cups,gripping palms and industrial cameras is designed;the industrial camera and robot arm consist of a hand-eye system,through the design of wireless camera communication scheme,hand-eye calibration and linear trajectory planning of the robot arm in Cartesian space,the robot can finish the archival box deposit and fetch task flexibly and coordinately.The software part is designed with a visual operation interface with debugging functions for each module and complete functions of archival box deposit and fetch,which provides real-time feedback on the current working status of the robot,realizing complete control of the robot in the general control room.Finally,experimental tests are conducted on the robot’s various module functions and complete deposit and fetch functions in the actual archive storage environment.First,by testing the SLAM mapping function and mobile barcode recognition function,combined with the actual navigation demonstration,the ability of navigation and addressing the archival box by the robot itself is proved,with the success rate of archival box finding reaching 98.67%.Second,comparison and ablation experiments on the two methods of precise positioning and measurement of archival boxes are conducted in terms of running time,correct detection rate,and positioning measurement accuracy,and the results show that the running time of both methods can be controlled within 1s,the correct detection rate can reach up to 96%,and the positioning and measurement error are within 2mm.At last,the actual deposit and fetch experiments show that the success rate of deposit and fetch archival boxes by the archival robot designed in this paper is more than 95%,which indicates the potential of practical application.