본문 바로가기
  • Home

CSRDCF_BD: Adaptive Spatial Mask Refinement for Visual Tracking via Bhattacharyya Distance

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(6), pp.119~129
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : May 15, 2025
  • Accepted : June 9, 2025
  • Published : June 30, 2025

Jung-Min Song 1

1국방과학연구소

Accredited

ABSTRACT

This paper proposes CSRDCF_BD, a novel extension of the CSRDCF tracker that addresses the limitations of spatial reliability modeling through adaptive filtering based on the Bhattacharyya Distance. The proposed method dynamically adjusts the weighting of binary masks by measuring histogram similarity between the foreground and background regions. This enables more robust filter learning under challenging visual conditions such as occlusion, background clutter, and illumination variation. Quantitative evaluations on the OTB100 and VisDrone benchmarks demonstrate that CSRDCF_BD consistently outperforms the baseline CSRDCF and other state-of-the-art trackers in terms of both precision and success rate. Furthermore, CSRDCF_BD exhibits superior temporal stability, as evidenced by its performance across attribute-specific and sequence-level analyses. These results confirm that refined spatial reliability modeling combined with adaptive filtering strategies can significantly enhance the robustness and generalization capability of correlation filter-based object tracking systems.

Citation status

* References for papers published after 2023 are currently being built.