查看文章

Vehicle type (make and model) recognition provides high level of security to the systems that are solely based on automatic license plate detection and recognition. Most of the work in this direction has been done in controlled conditions. In this paper we evaluate in an extensive experimental setting, the strength and weakness of various global and local feature based methods on vehicle images captured under controlled as well as uncontrolled conditions. We have introduced a challenging database that has been collected in complex conditions i-e scale, rotation, illumination variation, low contrast etc. Our method achieves 65 % rank-1 recognition accuracy on vehicle images captured under uncontrolled conditions with strong background clutter and 85 % recognition accuracy on segmented vehicle images captured under controlled conditions.