.png){kind=icon}
.png){kind=icon}
Return
GGSTrack: Geometric graph with spatio-temporal convolution for multi-object tracking
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131234.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Graph Convolutional Networks (GGCNs) for Multi-Object Tracking: The paper introduces a novel multi-object tracking algorithm, GGSTrack, which integrates geometric graph convolutional networks (GCNs) to model spatial relationships between objects. This approach enhances the tracking performance in dense scenes by capturing geometric interactions between targets. • Innovative Spatio-Temporal Modeling Framework: A spatio-temporal structure is proposed, combining geometric graph convolutions with temporal convolutional networks (TCNs) to comprehensively capture both spatial and temporal features of targets across multiple frames. This enables robust tracking in complex scenarios, such as occlusions and rapid motion. • Cascade Matching Framework: To improve tracking efficiency and precision, GGSTrack incorporates a four-stage cascade matching framework that progressively handles various tracking subtasks, including low-confidence candidate matching, occlusion handling, and duplicate detection. This modular design reduces computational complexity while maintaining high tracking accuracy. • Performance on Public Datasets: Extensive experiments demonstrate the superiority of GGSTrack across multiple benchmark datasets, including those with dense scenes and complex occlusion situations. The algorithm outperforms existing methods, setting a new standard in multi-object tracking. • New Temporal Convolutional Network (TCN) Application: The paper pioneers the use of Temporal Convolutional Networks in multi-object tracking, leveraging their ability to capture long-range temporal dependencies and enhance tracking stability, particularly in dynamic and challenging environments.
Keywords:
Graph Convolutional Networks
Multi-Object Tracking
Spatio-Temporal Modeling
Temporal Convolutional Networks
Cascade Matching
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W