Design And Implementation Of Lightweight Vehicle Detection And Tracking System Based On Deep Learning

With the continuous development of science and technology,artificial intelligence technology has also made a qualitative leap,and the field of computer vision in artificial intelligence has also occupied an important position.Among them,target detection and tracking are an indispensable part of the field of computer vision.Target detection and tracking have many practical applications in real life,including monitoring on traffic roads,face recognition of access control systems,biomedical diagnosis,etc.enumerate.Due to the continuous improvement of computing power and cloud technology,the deep learning technology that belongs to artificial intelligence has made great progress.Therefore,the target detection and tracking technology based on deep learning has also been continuously developed.Models based on deep learning are also improving in accuracy and speed.In the field of transportation,detecting and tracking vehicles in traffic bayonet surveillance video is a subject with practical needs and significance.Currently,there are different vehicle detectors and trackers in academic research and actual engineering,but these vehicle detectors and trackers are also difficult to meet the requirements of accuracy and speed.In view of the above problems,this thesis attempts to design a lightweight system for vehicle detection and tracking,which meets the requirements in terms of model size,detection and tracking speed and accuracy.The arrangement of this thesis is as follows: First,the background and significance of current target detection and tracking technology and related research status at home and abroad are described;second,the theoretical methods and specific processes of general lightweight vehicle detection and tracking technology are described;further,on this basis The general process is improved to try to achieve accuracy and real-time improvement.Specifically,in the target detection part,this thesis will introduce a deep learning method,use convolutional neural network for vehicle positioning,improve the accuracy of vehicle detection;at the same time,according to the monitoring video collected by the traffic bayonet The related processing is made into a self-built data set,and the neural network is trained to improve the generalization ability of the detection algorithm for different scenarios.In the vehicle tracking part,this thesis intends to try to use the Deep SORT tracking framework,use the deep learning method to train the re-identification network,and use the traditional tracking algorithm to speed up.On the basis of combining detection and tracking,use their respective characteristics to improve the accuracy of the model.Based on the above methods,continuous experiments and improvements are carried out,and the results obtained are compared and evaluated with the existing methods to achieve the goal of improving the existing technology.