Homogeneity patch search method for efficient vehicle color classification using front-of-vehicle image
Yoosoo Jeong, Kil Houm Park, Daejin Park · 2017
As one of the important features of a vehicle, color features are very discernible to classify the vehicle color using a front-of-the-vehicle image acquired from CCTV. The color identification of vehicles plays a significant role tracking crime vehicles in imaging systems based on CCTV. In this paper, we newly define a color homogeneity patch and propose its searching method. The proposed method focuses on reducing a intra-class variance by avoiding another color distribution, not a vehicle color, which is included in vehicle color features. In addition we could effectively improve this method based on a voting strategy. A region of interest (ROI) including bonnet is detected by predefined information about the given car and the search method is applied to ROI for selecting color homogeneity patches. The proposed approach extracts HSV histogram for each patch, and adopts the multi-adaboost algorithm to classify the color of each patches. We integrate the proposed method with the voting strategy to determine the color of the vehicle. For the validation of the feasibility of our approach, we compare the result with just a sliding window without consideration of homogeneity. Experiment results show that homogeneity patches search method can be efficiently applied to recognize vehicle color.