Vehicle identification based on feature point matching and epipolar geometry constraint

Vehicle identification has been extensively researched in recent years. Usually the license plate is used to identify the vehicle. However when the vehicle does not have a license plate or the license plate is occluded, other features of the vehicle have to be employed. In this paper, we propose a method to identify vehicles by feature point matching and epipolar geometry constraint. Affine-sift (Affine Scale-Invariant Feature Transform), which is invariant to affine transformation, is used to generate feature points, and a collection of matched feature point pairs is obtained which contains many false matched pairs. Then epipolar geometry constraint is employed to eliminate false matched point pairs. And the vehicle is identified by the number of corrected matched point pairs. Our method is motivated by the insight that same vehicle in different backgrounds has plenty of matched point pairs on the vehicle, and the matched point pairs should coincide with epipolar geometry constraint, while false matched point pairs in different backgrounds and on different vehicles will not coincide with epipolar geometry constraint. Experimental results show that our method is able to identify vehicles in different places under different view points.