Research On Vehicle Detection And Vehicle Recognition Based On Video Image

With the traffic accident and traffic congestion problem is increasingly serious,intelligent transportation system(ITS) has attracted more and more attention.Intelligent transportation system is the development trend of the future traffic, and its most important component is the vehicle detection and vehicle identification technology. Based on the domestic and foreign related research results, mainly studied the detection and classification of vehicle based on video image.Firstly, the foreground image of moving vehicle is obtained by using the Surendra background updating algorithm and the background difference method. Aiming at the problem that the fixed threshold cannot meet the need of real-time due to the change of the factors such as illumination, weather, shadow, the article adopts the method of dynamic threshold. Using particle swarm optimization algorithm combined with maximum entropy image segmentation method, obtain the optimal segmentation threshold. The gray level histogram and edge features are fused to eliminate vehicle shadow, and improve the accuracy of vehicle detection.Then, the geometric feature, texture feature and orientation gradient histogram(HOG)of the vehicle are extracted respectively. Features of each type were analyzed one by one,choose the perimeter, area and length to width ratio as a geometrical feature; mean,standard deviation, entropy, peace slide and consistency as texture features; hog as the local characteristics of the vehicle. Aiming at the problem that the HOG dimension is too high to calculate the complex problem, the HOG feature is reduced by the principal component analysis. Characteristics of the selected are combined into vehicle information,using the combination characteristics of the vehicle can effectively solve single feature susceptible to illumination, weather, shadows and other environmental impacts, help to improve the accuracy of vehicle recognition.Finally, using support vector machine classifier for cars, trucks and vans, which can identify three types of vehicle classification. According to the characteristics of three types of vehicle parameters to create sample database, using the grid search method, geneticalgorithm and particle swarm optimization algorithm respectively optimize the kernel parameters and punishment factor. The experimental results show that the classification accuracy of the vehicle is the highest when the parameters are optimized by using the particle swarm optimization algorithm.