Cascaded matching based on detection box area for multi-object tracking

被引:2
|
作者
Gu, Songbo [1 ]
Zhang, Miaohui [1 ]
Xiao, Qiyang [1 ]
Shi, Wentao [2 ]
机构
[1] Henan Univ, Sch Artificial Intelligence, Zhengzhou 450046, Peoples R China
[2] Northwestern Polytech Univ, Sch Marine Sci & Technol, Xian 710072, Peoples R China
关键词
Deep learning; Multi-object tracking; Cascaded matching; Detection box;
D O I
10.1016/j.knosys.2024.112075
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In the existing tracking-by-detection paradigm, advanced approaches rely on appearance features to establish associations between current detections and trajectories. However, these methods are often plagued by issues such as sluggish tracking performance and suboptimal results, particularly when confronted with the unreliability of the appearance features. Considering these challenges, we propose a novel cascaded matching algorithm called the detection box area-based tracking algorithm (DBAT), which groups the detection boxes by area size and associates detections within each group in a cascaded manner. To enhance the accuracy of grouping, we introduce two crucial components to enhance the quality of detections: the compressed self-decoding module (CSDM) and the task collaboration module (TCM). To acquire more precise location information and augment feature richness, CSDM decomposes the input features into two one-dimensional feature encodings and one two-dimensional feature encoding. Subsequently, these feature encodings perform feature aggregation along both spatial directions to capture long-range dependencies and refine the accuracy of location information. Ultimately, these aggregated features engage with the original features, facilitating information fusion and elevating the overall feature representation. To alleviate potential conflicts between various tasks and bolster task-specific representations, TCM combines disparate receptive fields and decouples features through self-relationship and cross-relationship mappings, thereby concurrently enhancing learning across different tasks. Extensive experiments demonstrate that our proposed method achieves performance comparable to state-of-the-art methods on the MOT17, MOT20 and DanceTrack benchmark tests.
引用
收藏
页数:11
相关论文
共 50 条
  • [31] Embedded Smart Tracker Based on Multi-object Tracking
    Xu Yan
    Wang Lei
    Liang Jianpeng
    Li Tao
    Cao Zuoliang
    THEORETICAL AND MATHEMATICAL FOUNDATIONS OF COMPUTER SCIENCE, 2011, 164 : 190 - 197
  • [32] Multi-Object Tracking Based on Segmentation and Collision Avoidance
    Meng Zhao
    Junhui Wang
    Maoyong Cao
    Peirui Bai
    Hongyan Gu
    Mingtao Pei
    Journal of Beijing Institute of Technology, 2018, 27 (02) : 213 - 219
  • [33] Multi-object tracking based on graph neural networks
    Zhang, Yubo
    Zheng, Liying
    Huang, Qingming
    MULTIMEDIA SYSTEMS, 2025, 31 (01)
  • [34] Pedestrian Multi-object Tracking Algorithm Based on Attention Feature Fusion
    Zhou, Yan
    Du, Zhennan
    Wang, Dongli
    ADVANCES IN COMPUTATIONAL INTELLIGENCE, IWANN 2023, PT I, 2023, 14134 : 105 - 118
  • [35] A hybrid optimisation enabled deep learning for object detection and multi-object tracking
    Thirumalai, J.
    Gomathi, M.
    Sindhu, T. S.
    Kumar, A. Senthil
    Puviarasi, R.
    INTERNATIONAL JOURNAL OF AD HOC AND UBIQUITOUS COMPUTING, 2024, 46 (03) : 150 - 165
  • [36] DOMOPT: A Detection-Based Online Multi-Object Pedestrian Tracking Network for Videos
    Huan, Ruohong
    Zheng, Shuaishuai
    Xie, Chaojie
    Chen, Peng
    Liang, Ronghua
    INTERNATIONAL JOURNAL OF PATTERN RECOGNITION AND ARTIFICIAL INTELLIGENCE, 2023, 37 (09)
  • [37] MULTI-OBJECT TRACKING AS ATTENTION MECHANISM
    Fukui, Hiroshi
    Miyagawa, Taiki
    Morishita, Yusuke
    2023 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING, ICIP, 2023, : 505 - 509
  • [38] Interacting Tracklets for Multi-Object Tracking
    Lan, Long
    Wang, Xinchao
    Zhang, Shiliang
    Tao, Dacheng
    Gao, Wen
    Huang, Thomas S.
    IEEE TRANSACTIONS ON IMAGE PROCESSING, 2018, 27 (09) : 4585 - 4597
  • [39] Multi-object tracking for horse racing
    Ng, Wing W. Y.
    Liu, Xuyu
    Yan, Xuli
    Tian, Xing
    Zhong, Cankun
    Kwong, Sam
    INFORMATION SCIENCES, 2023, 638
  • [40] Multi-object tracking algorithm based on multi-stage association
    Huo X.
    Gai S.
    Hong R.
    Zhou W.
    Da F.
    Yi Qi Yi Biao Xue Bao/Chinese Journal of Scientific Instrument, 2023, 44 (11): : 205 - 214