| 张全,刘田甜,刘洋毅,段昶,李艳,彭博.融合多尺度特征的雾天车辆重识别算法[J].电子测量与仪器学报,2026,40(1):269-278 |
| 融合多尺度特征的雾天车辆重识别算法 |
| Multi-scale feature fusion for foggy weather vehicle re-identification |
| |
| DOI： |
| 中文关键词: 车辆重识别 多尺度特征 雾天场景 特征融合 |
| 英文关键词:vehicle re-identification multi-scale features foggy scenes feature fusion |
| 基金项目:智能警务四川省重点实验室开放课题(ZNJW2024KFMS003)项目资助 |
|
|
| Author | Institution |
| Zhang Quan | School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu 610500, China |
| Liu Tiantian | School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu 610500, China |
| Liu Yangyi | Intelligent Policing Key Laboratory of Sichuan Province, Sichuan Police College, Luzhou 646000, China |
| Duan Chang | School of Electrical Engineering and Information, Southwest Petroleum University, Chengdu 610500, China |
| Li Yan | School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu 610500, China |
| Peng Bo | School of Computer Science and Software Engineering, Southwest Petroleum University, Chengdu 610500, China |
|
| 摘要点击次数: 585 |
| 全文下载次数: 501 |
| 中文摘要: |
| 车辆重识别技术在智能交通系统中扮演着至关重要的角色，其精准高效的性能对于显著提升城市交通安全性和交通效率具有决定性影响。然而，雾天等复杂天气条件会导致成像可见度降低，使得车辆外观信息严重失真，现有的车辆重识别方法仍然存在重识别平均精度较低和泛化能力不足的问题。为解决这些问题，提出了一种融合多尺度特征的雾天车辆重识别方法，旨在增强雾天真实数据下的重识别能力。该方法分为图像去雾和车辆重识别两个分支，通过共享权重的思想来平衡这两个任务，模型能够在雾天图像中提取稳定且具有代表性的特征。图像去雾模块采用两阶段恢复和金字塔增强技术生成清晰图像，以提供关键的雾天车辆图像特征，减小雾对重识别精度的影响。车辆重识别模块利用特征金字塔和卷积块注意力机制，从不同尺度上捕获更丰富和重要的特征，提升整个分支的重识别能力。在FVRID数据集上进行实验验证并与现有重识别方法进行对比，结果显示，在真实数据中的平均精度均值达到83.32%，首位命中率为94.70%，这两项指标均高于其他方法，这表明所提出的融合多尺度特征的雾天车辆重识别方法显著提升了雾天条件下真实数据的重识别性能，具有更强的准确性和泛化能力。 |
| 英文摘要: |
| Vehicle re-identification technology plays a crucial role in intelligent transportation systems. Its accurate and efficient performance is decisive for significantly enhancing urban traffic safety and efficiency. However, complex weather conditions such as fog can lead to reduced imaging visibility, severely distorting vehicle appearance information. Existing vehicle re-identification methods still suffer from lower average precision and inadequate generalization capabilities under these conditions. To address these issues, a method that integrates multi-scale features for vehicle re-identification in foggy weather has been proposed. This method aims to enhance the re-identification accuracy and robustness on real-world data under foggy conditions. This method is divided into two branches: image dehazing and vehicle Re-ID. By leveraging the concept of shared weights, this approach balances the two tasks, enabling the model to extract stable and representative features from foggy images. The image dehazing module utilizes a two-stage restoration and pyramid enhancement technique to generate clear images, providing key features of vehicles in foggy conditions, there-by reducing the impact of haze on the accuracy of Re-ID. The vehicle Re-ID module leverages a feature pyramid and convolutional block attention mechanism to capture richer and more significant features across different scales, enhancing the entire branch’s Re-ID capability. Experiments were conducted on the FVRID dataset, comparing this method with various other Re-ID approaches. The results showed that on real-world data, the mean average precision reached 83.32%, and the cumulative matching characteristic at rank 1 was 94.70%. Both metrics outperformed other methods, indicating that the proposed multi-scale feature fusion method for foggy weather vehicle Re-ID significantly improves performance under such conditions, demonstrating stronger accuracy and generalization capability. This research not only advances the current state of technology for foggy weather vehicle re-identification but also provides valuable insights for future studies. As the demand increases for applications such as intelligent traffic management and autonomous driving systems, this improved re-identification method holds great promise for advancing these related fields. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
