Transfer learning-based discriminative correlation filter for visual tracking
作者：
Highlights：
• We propose a novel Transfer Learning-based Discriminative Correlation Filter (TLDCF) to avoid the corruption of the updated filters and enhance the distinguishing ability of the model.
• We encode the spatio-temporal relationship between frames as Gaussian prior knowledge, which provides reliable cues for tracking and suppresses the significant location drift.
• We develop an efficient ADMM-based algorithm to calculate filters in the frequency domain in real time.
摘要
•We propose a novel Transfer Learning-based Discriminative Correlation Filter (TLDCF) to avoid the corruption of the updated filters and enhance the distinguishing ability of the model.•We encode the spatio-temporal relationship between frames as Gaussian prior knowledge, which provides reliable cues for tracking and suppresses the significant location drift.•We develop an efficient ADMM-based algorithm to calculate filters in the frequency domain in real time.
论文关键词：Visual tracking,Discriminative correlation filter,Instance-Transfer,Probability-Transfer
论文评审过程：Received 21 May 2019, Revised 30 November 2019, Accepted 10 December 2019, Available online 18 December 2019, Version of Record 26 December 2019.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.107157
