Vehicle Classification using Transferable Deep Neural Network Features

We address vehicle detection on rear view vehicle images captured from a distance along multi-lane highways, and vehicle classification using transferable features from Deep Neural Network. We address the following problems that are specific to our application: how to utilize dash lane markings to assist vehicle detection, what features are useful for classification on vehicle categories, and how to utilize Deep Neural Network when the size of the labelled data is limited. Experiment results suggest our approach outperforms other state-of-the-art.