Robust Classification of Vehicle based on Fusion of TSRP and Wavelet Fractal Signature

This paper presents a new vehicle's shape representation, which can describe its shape features. An algorithm called transformation-ring-projection (TRP), which is usually used in the recognition of characters in a binary image, is now applied to obtain multiple one-dimension patterns of vehicle shape. Firstly, in order to acquire high classification accuracy of vehicle types, we apply transformations-semi-ring-projection (TSRP) at eight central points which are distributed on the minimum ring of the vehicle region-of-interest (ROI) to traffic images and can obtain eight one-dimension patterns. Secondly, we calculate the fractal signatures in discrete wavelet transformation (DWT) domain of four one-dimension patterns. Thirdly, applying the MFC algorithm to process this kind of shape feature vector data set and generating several clusters. Finally, vehicles can be classified by shape feature matching system. Experiments results demonstrate the effectiveness and robustness of the proposed vehicle classification scheme.