Scale Decision Network for Correlation Filter-based Tracking

초록

In this paper, we present a new correlation-filter-based tracking framework with a scale-decision network to improve the performance of the correlation-filter-based tracking technique while preserving high-speed. The bounding box is fitted to the target using the relevant scale-changing actions determined by the scale-decision network. Because we can concentrate on a single-scale sample instead of the present approach's requirement for multi-scale sampling, our approach is computationally efficient. On the benchmark datasets, the proposed tracker outperforms all real-time trackers with a computation speed of over 100 frames per second. © 2022 IEEE.

키워드