Research On Vehicle Detection And Classification In Traffic Scene Images

With the increasing usage of vehicles and cameras,automated management of vehicles in traffic scene is becoming a major problem.The technology of vehicle detection and classification in traffic scene images is an important means to solve this problem.The selected topic has great theoretical and practical value.The main work is as follows:1.A training method of hidden variable part model for the vehicle was proposed.For vehicle detection,the hidden variable part model was trained for each class of vehicles.Based on hidden variable support vector machine,the model consists of three parts:the main model,the part model and the spatial relation of parts.The vehicle model can not only describe the appearance profile information of the vehicle as a whole,but also describe the part outline information of the vehicle.Experimental results show that the trained vehicle model can effectively detect the position of the vehicle in traffic scene images.2.A vehicle classification method based on vehicle hidden variable part model was proposed.The traffic scene images were detected by all of the vehicle models,and the vehicle image region was extracted by selecting the detection result of the maximum response.In the extracted vehicle image region,all of the models were used for model alignment to find the best positions of the main model and the part models.These positions can represent the characteristics of each vehicle class and reflect the unique information of the vehicle.The HOG features of all positions were extracted as the representation of the images and classified by SVM classifier.Experimental results show that compared with the existing methods,the proposed method has higher classification accuracy.3.A vehicle classification method based on convolutional neural network was proposed.The deep features of the main model and the part model were extracted by CNN,and the principal component analysis(PCA)was used to reduce the obtained high-dimensional deep features and then classified by SVM classifier.Experimental results show that this method can effectively improve the classification accuracy.