A Hierarchical Algorithm for Vehicle Model Type Recognition on Time-Sequence Road Images

This paper describes a vision-based algorithm for recognizing vehicle model types from time-sequence road images. Many types of vehicle models are offered commercially, and some of them resemble in shape. This prevents us to discriminate their model types from the others easily. To solve these problems, we propose a hierarchical recognition method with learning process, in which the resembling model groups are first generated and the effective features to discriminate the models in the each group are then selected using the subspace method in learning. In the recognition process, the front area of a vehicle is first detected from each frame of the input time-sequence images, then a hierarchical recognition which consists of a group and a category discrimination is performed. Finally, the results of frame recognition are integrated to realize stable recognition. The experimental results using time-sequence road images shows the proposed method is effective: the recognition rate for the registered model types is more than 99%, and the rejection rate for unregistered vehicle type is more than 92%