Vehicle classification based on multi-feature fusion

In this paper, we focus on the need for vehicle classification based on traffic surveillance videos. In the field of image classification, the representation of images can be realized in two ways, using global features and local features. Many methods using only global features can not cope with occlusion and spatial variations. Others based on local features and Bag-of-words models have been proved to be effective in solving above problems. However, global information plays an important role in vehicle classification and using local features alone performs poorly. In this paper, we present a method based on multi-feature fusion, which combines local feature and global feature together. After getting local information by SIFT, we extract global feature by means of PCA. Then we combine the features using a multiple kernel framework with a SVM classifier. We compare our approach with methods that use only local feature and global feature respectively on our dataset. Experimental results show that the proposed method performs better than the others.