Deep Feature Learning Network for Vehicle Retrieval

Vehicle Retrieval plays a very important role for traffic surveillance system which has been being developed in recent decades with innovative methods from traditional approach to deep learning approach. However, there are some challenges of Vehicle Retrieval which still need to be solved such as image resolution, image distortion, occlusion [1]. In this study, we utilized deep feature learning to build a vehicle retrieval system with two main tasks (1) deep feature learning network, (2) vehicle retrieval. Specifically, we adopted Faster R-CNN and SVM for building the feature network generating feature extraction and classification. Then, we used Approximate Nearest Neighbor (ANN) with tree-based approach for indexing feature vectors and searching. The proposed system solved some challenges such as dimensionality reduction of feature vector, improving the retrieval performance based on deep learning. The experimental results have shown that the proposed method is significantly outperforms the state of the art methods for vehicle retrieval on BIT-Vehicle dataset.