基于DeepSORT算法的前方道路车辆跟踪研究
Research on Vehicle Tracking on the Road Ahead Based on DeepSORT Algorithm
DOI: 10.12677/ORF.2023.136763, PDF,   
作者: 郭 薇, 王道斌, 李宸翔：武汉科技大学汽车与交通工程学院，湖北 武汉；朱泽德：安徽工业技术创新研究院六安院，安徽 六安
关键词: 道路车辆跟踪；DeepSORT；S2Net36；三元组损失；Road Vehicle Tracking； DeepSORT； S2Net36； Triplet Loss
摘要: 针对部分遮挡、目标漏检等导致的车辆ID频繁切换以及跟踪精度低的问题，本文提出了一种优化DeepSORT跟踪器的方法。一方面设计了S2Net36重识别网络：首先加深重识别网络构建ResNet36网络，提取更深层次的车辆外观特征；其次构建SER模块提取目标关键特征以及构建SE-Res2Net模块提取目标区域特征；最后基于ResNet36网络分别融合SER模块与SE-Res2Net模块得到S2Net36重识别网络。另一方面，引入三元组损失函数拉近相同目标不同样本之间的特征距离，通过提取更具有辨别力的车辆外观特征用于数据关联，进而提升对前方道路车辆的跟踪能力。实验结果表明，相比于DeepSORT原始算法，改进的算法的MOTA提高了1.18%，IDF1提升了0.80%，提高了对前方道路车辆的跟踪精度与稳定性，有望为自动驾驶车辆提供技术支持。
Abstract: This paper proposed a method to optimize the DeepSORT tracker aiming to solve the problems of frequent target ID switch and low tracking accuracy caused by target missed detection and occlusion. Firstly, a re-identification network called S2Net36 was designed to extract deeper vehicle appearance features. To achieve this, the re-identification network is deepened to build the ResNet36 network. Then, a SER module is constructed to extract the key features of targets and a SE-Res2Net module is constructed to extract the regional features of targets. Finally, the SER module and the SE-Res2Net module were embedded in the ResNet36 network to obtain the S2Net36 re-identification network. Secondly, a triplet loss function was introduced to shorten the feature distance of different samples of the same target. Extracting more discriminating vehicle appearance features for data correlation to improve the tracking ability for the vehicles ahead. The experimental results show that the proposed algorithm could improve the MOTA by 1.18% and IDF1 by 0.80% compared with the original DeepSORT algorithm. This improvement in tracking accuracy and stability of road vehicles ahead is expected to provide technical support for autonomous vehicles.
文章引用：郭薇, 朱泽德, 王道斌, 李宸翔. 基于DeepSORT算法的前方道路车辆跟踪研究[J]. 运筹与模糊学, 2023, 13(6): 7806-7816. https://doi.org/10.12677/ORF.2023.136763

参考文献

[1] 伍瀚, 等. 基于深度学习的视觉多目标跟踪研究综述[J]. 计算机科学, 2023, 50(4): 77-87.
[2] 贺愉婷, 车进, 吴金蔓. 基于YOLOv5和重识别的行人多目标跟踪方法[J]. 液晶与显示, 2022, 37(7): 880-890.
[3] 储琪. 基于深度学习的视频多目标跟踪算法研究[D]: [博士学位论文]. 合肥: 中国科学技术大学, 2019.
[4] Bewley, A., Ge, Z., Ott, L., et al. (2016) Simple Online and Realtime Tracking. 2016 IEEE International Conference on Image Processing (ICIP), Phoenix, 25-28 September 2016, 3464-3468. [Google Scholar] [CrossRef] 
[5] Wojke, N., Bewley, A. and Paulus, D. (2017) Simple Online and Realtime Tracking with a Deep Association Metric. 2017 IEEE International Conference on Image Processing (ICIP), Beijing, 17-20 September 2017, 3645-3649. [Google Scholar] [CrossRef] 
[6] 殷远齐. 基于机器学习的前方车辆行为识别方法研究[D]: [硕士学位论文]. 西安: 长安大学, 2022.
[7] Bochkovskiy, A., Wang, C. and Liao, H.M. (2020) YOLOv4: Op-timal Speed and Accuracy of Object Detection. https://arxiv.org/pdf/2004.10934.pdf
[8] Woo, S., Park, J., Lee, J.Y., et al. (2018) Cbam: Convolutional Block Attention Module. In: Ferrari, V., Hebert, M., Sminchisescu, C., Weiss, Y., Eds., Proceedings of the European Conference on Computer Vision (ECCV). Springer, Cham, 3-19. [Google Scholar] [CrossRef] 
[9] Gao, S.H., Cheng, M.M., Zhao, K., et al. (2019) Res2Net: A New Multi-Scale Backbone Architecture. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43, 652-662. [Google Scholar] [CrossRef] 
[10] 顾立鹏, 等. 无人车驾驶场景下的多目标车辆与行人跟踪算法[J]. 小型微型计算机系统, 2021, 42(3): 542-549.
[11] Hu, J., Li, S., Gang, S., et al. (2018) Squeeze-and-Excitation Networks. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, 18-23 June 2018, 7132-7141. [Google Scholar] [CrossRef] 
[12] Zhou, X., Wang, D. and Krähenbühl, P. (2019) Objects as Points. arXiv preprint, arXiv:1904.07850.
[13] 何维堃, 彭育辉, 黄炜, 等. 基于DeepSort的动态车辆多目标跟踪方法研究[J/OL]. 汽车技术: 1-7. 2023-09-20.[CrossRef] 
[14] 尤晓雨. 基于改进的YOLOv5和DeepSort车辆检测跟踪算法研究[D]: [硕士学位论文]. 西安: 长安大学, 2022.
[15] Wang, H., Zhang, F. and Wang, L. (2020) Fruit Classification Model Based on Improved Darknet53 Convolutional Neural Network. 2020 International Conference on Intelligent Transportation, Big Data & Smart City (ICITBS), Vientiane, 11-12 January 2020, 881-884. [Google Scholar] [CrossRef] 
[16] 金立生, 华强, 郭柏苍, 等. 基于优化DeepSORT的前方车辆多目标跟踪[J]. 浙江大学学报(工学版), 2021, 55(6): 1056-1064.
[17] Schroff, F., Kalenichenko, D. and Philbin, J. (2015) Facenet: A Unified Embedding for Face Recognition and Clustering. Pro-ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Boston, 7-12 June 2015, 815-823. [Google Scholar] [CrossRef] 
[18] Liu, X., Liu, W., Mei, T., et al. (2016) A Deep Learning-Based Approach to Progressive Vehicle Re-Identification for Urban Surveillance. In: Leibe, B., Matas, J., Sebe, N., Welling, M., Eds., European Conference on Computer Vision. Springer, Cham, 869-884. [Google Scholar] [CrossRef] 

为你推荐