Multi-View Feature Fusion Network for Vehicle Re-Identification


Haoran Wu, Dong Li, Yucan Zhou and Qinghua Hu, Tianjin University, China


Identifying whether two vehicles in different images are same or not is called vehicle reidentification. In cities, there are lots of cameras, but cameras cannot cover all the areas. If we can re-identify a car disappearing from one camera and appearing in another in two adjacent regions, we can easily track the vehicle and use the information help with traffic management. In this paper, we propose a two-branch deep learning model. This model extracts two kinds of features for each vehicle. The first one is license plate feature and the other is the global feature of the vehicle. Then the two kinds of features are fused together with a weight learned by the network. After, the Euclidean distance is used to calculate the distance between features of different inputs. Finally, we can re-identify vehicles according to their distance. We conduct some experiments to validate the effectiveness of the proposed model.


vehicle re-identification, deep learning model, feature fusion

Full Text  Volume 7, Number 13