Pyramid Histogram of Oriented Gradient and Particles Swarm Optimization Based SVM for Vehicle Detection

Vehicle Detection is an important part in intelligent transportation system (ITS) and driver assistance system. Considering vehicles have strong edges and lines in different orientation and scales, in this paper, we presents a method for detecting vehicles based on a feature named Pyramid Histogram of Oriented Gradient. This feature provides spatial distribution information of edges which was often ignored by other features. Specifically, we extract PHOG features from a traffic image and a vector is obtained as representation of this image. In order to speed up the process of calculation, Principle Component Analysis (PCA) algorithm is applied to these vectors to reduce their dimensionality. Tests show the efficiency improved significantly. Theses representative vectors are then used to train SVM classifier. To ensure the final classification accuracy, we adopt Particles Swarm Optimization (PSO) method to gain the parameters used in SVM classification. Optimal parameters correspond with optimal result. Experiments demonstrate the superiority of the proposed approach which has achieved an average accuracy of 95% on test images.