Research On Technology Of Logistics AGV Navigation Based On Vision

The rapid development of domestic manufacturing and logistics industries promotes the development of automated transportation tools.AGV(Automated Guided Vehicle)is a material transport vehicle with flexible,intelligent features and a high degree of automation.It has an important role that cannot be ignored in many fields.Positioning and navigation is the core technology of AGV,which determines the performance,intelligence and practicability of AGV.This dissertation studies the navigation scheme of AGV based on the system design idea of software-defined products.The research breaks the limitation of physical space,considers the security and stability of the system operation,and builds a visual auxiliary subsystem for the AGV system in the workshop logistics scene.This makes the navigation of AGV system to adapt to the needs of frequent restructuring and reconstruction of logistics sites,so as to build low-cost and high-reliable navigation solutions with higher adaptability to the environment.This dissertation takes the AGV system in the logistics scene as the research object,and briefly describes the main components of the AGV system and its working principle,and analyzes the requirements of the application scenario.Starting from low cost and high reliability,the application of AGV system focuses on the global positioning of visual non-global field of view environment and the intelligent detection function of AGV real-time vehicle vision,and builds an AGV overall solution based on the cloud-edge-end architecture.According to the work objectives,the function and equipment selection of the visual assistance system are determined,and the design and test verification are carried out for the construction of the system communication scheme.Aiming at the recognition problem of the AGV in the running state when the camera is located in the global perspective and the problem of positioning through visual technology,it is first converted into the problem of dynamic target detection in a static background,and the method of vision-based motion estimation is analyzed from a theoretical point of view.The RTSP protocol is used to obtain video images,and an improved adaptive thresholding and hole filling inter-frame difference algorithm based on OSTU is proposed to complete the foreground extraction and improve the accuracy of dynamic target detection.It lays a good foundation for the effectiveness of subsequent positioning algorithms.In order to realize the localization function of the AGV visual assistance system,based on the AGV foreground extraction algorithm in the running state,based on the extensive prior knowledge of the AGV operation scenario,this dissertation studies the localization algorithm from the global perspective.According to the imaging model and distortion correction model of the camera,the internal parameter calibration of the camera is completed by Zhang’s calibration method.For the Pn P coordinate transformation problem,an absolute positioning algorithm based on Harris corner detection and DLT solution based on calibration checkerboard is proposed,which improves the uniqueness of the solution of visual positioning results,and finally calculates the global positioning coordinates.And for the recognition of AGV number,an algorithm based on BP neural network for character recognition of simple projection patterns is proposed.Through the above steps,the interface development and functional implementation of the visual positioning system are performed to verify the effectiveness,accuracy and reliability of the positioning algorithm.In addition,a simple integrated navigation solution that combines motion control and machine vision is designed,and a theoretical analysis of the fusion positioning problem is carried out from the perspective of probability,which proves the rationality of the solution.Finally,the derivative algorithm of the Kalman filter algorithm is used.EKF completes the algorithm simulation of fusion positioning.Aiming at the problem of running posture recognition when AGV docks with important stations,artificial road signs are designed;at the same time,considering the conflict between AGV and obstacles when running,a deep learning-based classification algorithm is used to detect road signs and obstacles.In this dissertation,visual equipment is used to collect images of relevant obstacles and artificial signs and use a unified format as a training set,and the YOLOv2 network model is used to train the collected data set.In addition,for the recording problem of AGV transmission operation process,the QR code identification is added to the station,and then the current AGV operation node is identified by the on-board visual device.The experimental results show that the vehicle-mounted visual assistance system designed in this dissertation can realize manual identification,obstacles and QR code identification algorithm,and have certain effectiveness and real-time.