Edge Computing For Vision Detection In AGV Path Navigation

This paper addresses the problem of long computation time for autonomous navigation of AGVs in the traditional cloud service model,which cannot meet the computation of obstacle avoidance for sudden obstacles.In the face of map modeling,path planning and edge computing,we propose an AGV cart model with flexibility and real-time autonomous obstacle avoidance.The main work of this paper has the following three points:(1)The first part is the map building part,and the main research contents of this part are: analyzing SLAM simultaneous localization and map building techniques,and using visual SLAM system to build sparse point cloud maps of the environment.After that,local adaptive thresholding is introduced for the disadvantage of weak feature point extraction,and octree map construction method and projection along z-axis are introduced for the disadvantage that sparse point cloud map cannot be navigated directly,which converts the original point cloud map into 2D raster map and ensures the smooth operation of subsequent autonomous path planning and navigation of AGV.(2)The second part is the path planning part,which analyzes the characteristics of the global path planning algorithm.To address the problems of decreasing population diversity,falling into local optimum and slow convergence at the end of iteration of the sparrow search path planning algorithm selected for global path planning in this topic.The Tent mapping is introduced to initialize the individual sparrow positions and increase the diversity;the previous generation global optimal solution is introduced in the discoverer position update formula as well as the adaptive weight ω is added to improve the adequacy of global search and the ability to coordinate local search and global exploration.(3)The edge computing part addresses the situation that AGV trolleys take a long time and do not respond timely through traditional obstacle avoidance calculation methods when encountering sudden unknown obstacles.Based on ROS(Robot Operating System),which provides various functions such as positioning,modeling,and navigation,edge computing nodes are introduced into the AGV cart,and the AGV can offload computing tasks to the edge nodes to solve the problem of insufficient computing resources.