VEHICLE CLASS RECOGNITION USING 3D CG

Abstract

This paper describes a robust method for recognizing vehicle classes. We have already developed vehicle recognition system based on local-feature configuration. The algorithm is based on our previous work, which is a generalization of the eigen-window method. It is one of image-based systems, hence, we usually need a lot of training images. But our system still works on the training images made from 3-dimentional computer graphic(CG) model vehicles. In our previous work, we have confirmed that our system can recognize one vehicle class very well, but we have realized that our system does not work very well on classifying many classes. In this paper, we describe the improvement of our recognition system to distinguish similar classes. As a result, our new system can recognize vehicle class using small number of CG training images. In other words, we can recognize vehicle classes without a time consuming and hard task of collecting real training images. We have confirmed this fact, performing outdoor experiments. Our system can recognize four vehicle classes of sedan, wagon, mini-van and hatchback from real images with accuracy of over 80%.