VEHICLE MAKE AND MODEL RECOGNITION USING LOCAL FEATURES AND LOGO DETECTION

Vehicle Make and Model Recognition (VMMR) plays an important role in Intelligent Transportation Systems. Due to the increase in volume and diversity of vehicles on the road, traffic monitoring has become both important and difficult. Several cues could be used in VMMR. Examples include salient features extracted from vehicle representing the shape of different parts or logo detection and recognition. The challenge is in identification of regions of interest and optimal fusion of pertinent cues. This paper presents a two-stage framework for VMMR that uses Multiple Instance Learning for fine grained classification. Our solution lies in constraining the high dimensional instance space by selecting the most informative clues of a given visual category. In the second phase, vehicle manufacturer logo is detected for the images classified with low confidence in the first stage, to verify the output of system. Empirical evaluation on a very large VMMR dataset confirms the effectiveness of the proposed scheme.