Design And Implementation Of Surveillance Video Vehicle Classification System Based On CNN

With the development of artificial intelligence,more and more researchers are working on the Intelligent Traffic System(ITS).The aim of Intelligent Traffic System(ITS)is to build an efficient traffic system to manage the various traffic problems.In this thesis,we mainly focused on the highway surveillance video vehicle recognition system which is an important part of Intelligent Traffic System(ITS).Inspired with the successful of deep neural networks,the Convolution Neural Networks(CNN)are utilized to improve the performance.From the extensive experiment,a most recent Convolution Neural Networks,Residual Network(ResNet)is used to conduct the vehicle classification.In our procedures,we first extract the vehicle images of surveillance video in which the background subtraction method is used to detect moving object.After the detection,to improve the performance,we also implement the vehicle tracking system.Then the Residual Networks(ResNet)is used to conduct the classification of the vehicle.Based on the algorithm,we also build a distributed system,named Celery,which utilizes the distributed system to share the large-scale computing burden.With this distributed system,the various real surveillance video vehicle systems are implemented by parallel and multi-thread way,which can handle the real highway surveillance vehicle recognition task.Finally the Celery system analyzes and presented the result in the form of charts.The information of charts includes the number of different vehicle type of highway surveillance video,the changing number of different vehicle type of highway surveillance video in different time.These statistics information will help the user know about traffic information in time.