Deformable 3-D model based vehicle matching with weighted Hausdorff and EDA in traffic surveillance

Abstract

3-D model based objects matching is a fundamental in image processing and computer vision, especially for object localization, tracking, and recognition. In this paper a new deformable models with commonly 9–12 length-angle shape parameters are used for matching, which can represent rich shape details for traffic vehicle classification. A Weighted Modified Square Haudsorff Distance (WMSHD) is designed to suppress the noise brought by the low quality of object in feature extraction, and then the weight is defined in the models and object edge points. Estimation of Distribution Algorithm (EDA) is used as the evolutionary algorithm to search the best parameters of model shape and localization with shape parameters and pose parameters. Experiments are made with histograms, curves. The matching results show that the proposed method is effective.