Model-based Vehicle Make and Model Recognition from Roads

Abstract

In computer vision, the vehicle detection and identification is a very popular research topic. The intelligent vehicle detection application must first be able to detect ROI (Region of Interest) of vehicle exactly in order to obtain the vehicle-related information. This paper uses symmetrical SURF descriptor which enhances the ability of SURF to detect all possible symmetrical matching pairs for vehicle detection and analysis. Each vehicle can be found accurately and efficiently by the matching results even though only single image without using any motion features. This detection scheme has a main advantages that no need using background subtraction method. After that, modified vehicle make and model recognition (MMR) scheme has been presented to resolve vehicle identification process. We adopt a grid division scheme to construct some weak vehicle classifier and then combine such weak classifier into a stronger vehicle classifier. The ensemble classifier can accurately recognize each type vehicle. Experimental results prove the superiorities of our method in vehicle MMR.