The Research On The Algorithm Of Video-based Vehicle Detection

With the rapid development of economy,people's living standards are constantly improving,the use of global cars is also growing fast.However the traffic situation is getting worse and the problem of traffic congestion has become a common phenomenon,those issues has caused a lot of trouble to people's life and travel.Traffic accidents are also frequent.So the traffic problem has become the focus of attention of all around the world.Now,there are two main ways to solve the traffic problem,one is to accelerate the improvement of the urban transport system,for example widen roads and build viaducts.These methods can alleviate the traffic jam but they need to invest a lot of manpower,material resources and financial resources.Obviously not the ideal solution.In order to completely solve the traffic problem,another is to apply science and technology,such as computer technology,image processing technology,information and communication technology and pattern recognition technology,to the increasingly serious traffic problems.Therefore,the Intelligent Transportation System comes into being.Vehicle detection as a key component of intelligent transportation system,it plays an important role in intelligent transportation system.Many important traffic parameters can be obtained by vehicle detection technology,those parameters provides important basis for the traffic monitoring and management.In this paper,vehicle detection in video is studied and analyzed.The main research of this paper includes Extraction of Vehicle Image Features,Vehicle Classification Model Training,Acquiring ROI(the region of interest),Video-Based Vehicle Detection and Experimental Conclusion Analysis.In part of extraction vehicle image feature we extract three different feature of vehicle image,vehicle image features based on SIFT algorithm and Bag-of-Feature algorithm,vehicle image features based on SURF algorithm and Bag-of-Feature algorithm and vehicle image features based on HOG algorithm.The SVM algorithm is used as a classifier for the training model in the vehicle classification model training section,in the process of training model,a large number of parameter are adjusted and kernel functions are selected to obtain the optimal model.In part of Acquiring interest area,the method of frame difference method and gaussian mixture model is used.In part of Video-Based Vehicle Detection,vehicle classification model was applied to ROI,the area of identified as the vehicle is marked with a rectangular box.In the last part,we analyzed the advantages and disadvantages of the Video-Based Vehicle Detection model and proposed some improvements.