Vehicle matching using color

We describe a vehicle matching algorithm, which is part of our AutoColor traffic monitoring system, that can identify the same vehicle at different camera sites using color information only. This work is challenging since vehicle color varies with daylight illumination which is unpredictable. We present a novel color matching approach that takes advantage of road color variation to model changes in illumination. This model is used to compensate vehicle color variation and minimize the false positive matches. The system is tested on real images with promising results in comparison with other color matching algorithms.