Vehicle Re-Identification Based on Multi-View Fusion of Positions
Xing Xu, Xiangbin Shi · 2023
Vehicle re-identification encounters two main challenges, including significant appearance variations of vehicles from different viewpoints and the mismatches between extracted features and true features on illumination, occlusion, and low resolution. To address these challenges, we propose a network based on multi-position viewpoint fusion. Firstly, the ConvNeXt network is utilized as the backbone, including response normalization modules to learn all salient local features within each viewpoint adaptively. Secondly, an embedding layer based on multi-position viewpoint information is proposed to extract more details and invariant features from the images. To accelerate the convergence speed of the model, batch normalization is utilized to constrain the features and minimize differences in optimization across various loss functions. We conduct experiments on two public datasets, VeRi-776 and VehicleID, and the results demonstrate the effectiveness of our proposed method in the vehicle re-identification task.