A rapid vehicle recognition and retrieval system

Abstract

This paper presents a fast vehicle recognition and vehicle retrieval system based on “bag of words”. In this system, the input is an image of vehicle and the vehicle will be identified automatically, it can also retrieve images which are similar to the input image. 3742 vehicle images which include 28 types of vehicles are collected as the image database. Features of these images are extracted and quantized as “visual words” which are clustered so that every vehicle model is represented by a set of “visual words”. A hessian-affine detector is used to detect the features of the input image and the SIFT descriptor transforms each feature into a 128-dimensional vector which is finally quantized as a “visual word”. In this way, the input image is transformed into a set of “visual words”. By matching the “visual words” between the vehicle model and the input image, it can compute the similarity and judge the type of the input vehicle image. For the vehicle retrieval, the “visual words” is weighted in the method of “TF-IDF” to compute the cosine similarity of two images. The retrieval results are ranked by cosine similarity. The experiments show that the precision of vehicle recognition is above 85%, and the using of “bag of words” is much faster than the traditional features matching.