Research On Vehicle Recognition Algorithm Based On Edge Intelligence

With the development of transportation infrastructure,urban traffic pressure has increased dramatically.In order to alleviate traffic congestion,many cities have installed thousands of traffic monitoring equipment in urban areas.However,since surveillance cameras shoot videos with a fixed angle of view,and these videos are always transmitted to the monitoring center,most of them are manually analyzed,which is expensive and the work effectiveness is low.Driven by this trend,there is an urgent need to push vehicle detection technology to the edge of the network,share the pressure on the core of the network,and improve the efficiency of network traffic control.In order to meet this demand,drones are used for transportation assistance operations,and edge computing and deep learning are combined for drone-side deployment.The high-resolution,high-frame-number,and large-volume videos taken by drones It adopts a filter method for preprocessing,and at the same time,selective model compression is carried out for the characteristics of large size,high accuracy,high delay and small size,low accuracy,and low delay of the target detection model based on deep learning,so that the vehicle detection model can be suitably deployed on the UAV side,and the vehicle can be detected and the delay and accuracy can be balanced.Therefore,this is a work worthy of in-depth research and development prospects.This article mainly reviews the background and motivation of edge intelligence operating at the edge of the network.At the same time,it analyzes the current research status of target detection algorithms and video preprocessing,understands some of the current challenges in the field of edge intelligence and target detection algorithms,and addresses these the challenge proposes a solution,the main innovations are as follows:(1)In order to reduce costs and improve efficiency,the transportation department currently uses drones to assist operations.Aiming at the problem of a large number of redundant video frames in vehicle videos taken by UAVs and the difficulty of deploying high-precision target detection models in embedded devices,this paper proposes a vehicle target detection method for UAV video analysis.This method sets a two-stage filter on the edge device to filter a large number of redundant frames through the pixel-level and structural differences of the frame,thereby greatly reducing the number of frames of the detection model transmitted to the back end;at the same time,channel pruning and layer pruning are adopted.The combined method compresses the YOLOv3 model and deploys it on the PC to achieve a balance between delay and accuracy.(2)In order to make full use of edge equipment resources and better improve the detection effect,this paper proposes a vehicle target detection method based on edge intelligence.This method improves the manual compression model to a compression model based on reinforcement learning,avoids manual parameter adjustment,and optimizes the efficiency of model compression.Deploy a video preprocessing mechanism and an automated compression deep learning model on embedded devices,and deploy a high-precision target detection model trained through migration learning in the cloud.Experimental results show that the proposed method has high accuracy when deployed on embedded devices and lower latency.