Vehicle Re-Identification for Intelligent Transport System Using DOLGNet Attention Model

Vehicle re-identification is an important task that involves identifying a specific vehicle from multiple images in computer vision. In recent years, deep learning-based approaches have shown promising results in this field. However, due to the significant variations in vehicle appearances caused by different viewpoints, occlusions, and lighting conditions, it remains a challenging problem. This paper proposed a Deep Orthogonal Local and Global Network (DOLGNet-Attention) for vehicle re-identification, which uses ResNet as the backbone and integrates self-attention and dilated convolution layers to effectively capture local and global features. Our approach combines local and global features by incorporating self-attention and dilated convolution layers into the network architecture. The self-attention mechanism focuses on relevant components within the local feature maps. At the same time, the dilated convolution layers capture information from multiple scales, which enhances the network's ability to handle variations in vehicle appearance. The results show that DOLGNet-Attention achieves top performance in terms of accuracy and robustness. Testing and training accuracy are done to determine the accuracy of the model. It determines image matches and non-match images compared to query images. Similarity values greater than 90% indicate that the image is similar to the query and the vehicle is re-identified. The DOLGNet-Attention model represents a significant advancement in vehicle re-identification, with the potential to improve the accuracy and efficiency of applications such as traffic management, surveillance, and security.