Vehicle detection with a part-based model for complex traffic conditions

In complex urban traffic conditions, occlusion between vehicles is a common problem which is challenging to current vehicle detection methods. In this paper, we have proposed a vehicle detection method based on a part-based model which can deal with the occlusion problem. Our method includes two steps: constructing the part-based model and detecting vehicles from traffic images. In the first step, a vehicle is divided into two parts representing an easily-occluded region around license plate and a commonly-visible region around vehicle window. Each part has low intra-class difference and is modeled by hybrid image template (HIT) with multiple types of feature descriptors in this paper. These two parts constitute our part-based model which is beneficial to vehicle detection with occlusion because the occlusion of one part has no impact on the detection of the other part. In the second step, we detect vehicles from the input image. The detection process first identifies the part candidates by using template matching and then combines the part candidates for detecting vehicles. To test our method, we have done several experiments on complex urban traffic conditions with occlusions. The experimental results show that our method can effectively cope with partial occlusion. Moreover, our method can also adapt in slight vehicle deformation and different weather conditions.