Vehicle size classification for real time intelligent transportation system

This paper proposes a vehicle size classification system which distinguishes small-size cars, medium-size cars, and big-size cars automatically. Previous vehicle size classification researches usually fixed the camera viewpoint or limited it to small orientations. The proposed system utilizes the concavity property of sedans and buses to distinguish small-size and big-size cars in large orientations. Then, the head width of a car is estimated and two aspect ratios of car height to head width and head width to car width are designed. The multiclass support vector machine (SVM) is adopted to classify the aspect ratios of cars into different sizes. In the experiments, various kinds of color and model of car images are collected, and satisfactory size classification accuracy is proved to be provided by the proposed algorithm. Furthermore, the computation time of the entire system is less than 0.01 second per image; therefore, the proposed method is applicable for real-time intelligent transportation systems.