A Vehicle Re-Identification Model Based on Global and Local Feature Enhancement and Fusion

Vehicle re-identification seeks to identify all images of vehicles with the same identity as a queried vehicle in a large database. Current algorithms face challenges of low accuracy and model redundancy due to varying vehicle appearances from different angles and minimal differences among different vehicles in specific views. This paper proposes an end-to-end dual-branch vehicle re-identification model with both global and local feature enhancements. The network aims to boost the representation capability of vehicle features and obtain more comprehensive and distinctive vehicle characteristics through the fusion of dual-branch features. The model is divided into two branches representing global and local features respectively. A spatial scale enhancement module is employed in the global feature branch to enhance the perception of spatial information of vehicles. In the local branch, a strategy of feature map segmentation in the middle layers of the network is utilized to encourage the model to explore more crucial local details. The model is trained using a combination of two loss functions to enhance the recognition capability of the network.Experimental results demonstrate that the proposed model achieves an mAP of 84.5% and CMC@1 of 97.6% on the VeRi-776 dataset. On the VehicleID dataset, the CMC@1 values for the three sub-datasets are 87.2%, 84.6%, and 81.3%, respectively, surpassing advanced algorithms such as PVEN, SSBVER, and TANet.The research findings affirm the advancement and effectiveness of the proposed model.