.png){kind=icon}
.png){kind=icon}
Return
Learning sparse spatial attribute-aware correlation filter tracking via rank-based surrounding strategy
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-025-01890-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Discriminative correlation filters have shown promising results in object tracking by using spatial and temporal regularization to reduce boundary effects and guide model updates. However, traditional approaches often apply uniform regularization across multi-channel features, limiting their adaptability to appearance changes and complex backgrounds. This study introduces the surrounding sparse spatial attribute-aware correlation filter (SSSACF) to address these challenges. The method incorporates attribute-aware sparsity to extract meaningful information from multi-channel attribute patterns, enhancing object discrimination. A novel ranking mechanism samples adjacent patches around the target, integrating contextual information for improved localization and continuous tracking in dynamic scenes. Furthermore, a sparse spatial integration method effectively suppresses adverse background influences and boundary effects using reference weights. The proposed SSSACF tracker achieves robust and efficient tracking, significantly outperforming state-of-the-art methods on publicly available datasets, demonstrating its effectiveness in addressing the limitations of traditional correlation filters.
Keywords:
Convolutional neural networks
Sparse spatial information
Attribute-aware
Rank-based surrounding information
Correlation filters
Visual object tracking
Journal
IF:
3.1
Papers:
2.7K
Citations:
2.7K