VRSDNet: vehicle re-identification with a shortly and densely connected convolutional neural network

Abstract

Vehicle re-identification aiming to match vehicle images captured by different cameras plays an important role in video surveillance for public security. In this paper, we solve Vehicle Re-identification with a Shortly and Densely connected convolutional neural Network (VRSDNet). The proposed VRSDNet mainly consists of a list of short and dense units (SDUs), necessary pooling and spatial normalization layers. Specifically, each SDU contains a short list of densely connected convolutional layers and each convolutional layer is of the same appropriate channels. As a result, the number of connections and the input channel of each convolutional layer are restricted in each SDU, and the architecture of VRSDNet is simple. Extensive experiments on both VeRi and VehicleID datasets show that the proposed VRSDNet is obviously superior to multiple state-of-the-art vehicle re-identification methods in terms of accuracy and speed.