A Two-stage Vehicle Type Recognition Method

Vehicle type recognition is a common question in modern intelligent transportation system. Although many methods have been proposed in literatures, how to recognize the vehicle type in the case of small samples is still troubling. To solve the problem mentioned above, this paper presents a method for vehicle type recognition based on a two-stage strategy that includes data preprocessing phrase and training phrase. The main purpose of the first stage is to remove the background of the vehicle image to reduce the interference of redundant information on subsequent stage. In second stage, a concept of multi-scale fusion feature which integrates handcrafted features with learning-based feature is to put forward to describe the characteristics of vehicles. The combined features are input to Support Vector Machine with Racial Basis Function (RBF-SVM) to train the recognition model. The proposed method is tested on MVVTR and the recognition accuracy is up to 90.96%, which is better than other methods. It is verified that strong generalization ability can be achieved in the case of small samples by using the twostage strategy.