Vehicle Classification with Convolutional Neural Network on Motion Blurred Images

Abstract

Vehicle classification is an important research area of Intelligent Traffic System. We first collect large set of static and clear vehicle images from internet and split the images into train and validation sets. Then we train the dataset with convolutional neural network (CNN). Experimental results show that CNN achieve high accuracy on the validation dataset. However, the trained model achieves poor performance on motion blurred images captured from videos. This paper proposes a new method for dealing with motion blurred images. Random blurred images are generated during training in order to optimize the network parameters. The final experimental results show that our proposed method achieves better performance than training directly with CNN.