Multi-level context-adaptive correlation tracking

作者：

Highlights：

• We propose the context pyramid to use multi-level contexts.

• The 3D spatial window can construct different levels with extracting feature once.

• Our method learns adaptive weights for different levels in the DCF.

• With hand-crafted features, we achieve comparable results to deep learning methods.

摘要

•We propose the context pyramid to use multi-level contexts.•The 3D spatial window can construct different levels with extracting feature once.•Our method learns adaptive weights for different levels in the DCF.•With hand-crafted features, we achieve comparable results to deep learning methods.

论文关键词：Visual tracking,Correlation filter,Context-adaptive tracker,Context pyramid

论文评审过程：Received 13 March 2018, Accepted 9 October 2018, Available online 11 October 2018, Version of Record 25 October 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.10.013