Deep asymmetric video-based person re-identification

作者：

Highlights：

• We address the “view-bias” problem, a key challenge of video-based person re-id.

• We propose a Deep Asymmetric Metric learning (DAM) method that embeds an asymmetric metric into a deep neural network.

• To make DAM scalable to large amount of camera views, we develop a clustering-based DAM.

• Extensive evaluations on three video-based person re-id datasets have shown the effectiveness of our asymmetric modelling.

摘要

•We address the “view-bias” problem, a key challenge of video-based person re-id.•We propose a Deep Asymmetric Metric learning (DAM) method that embeds an asymmetric metric into a deep neural network.•To make DAM scalable to large amount of camera views, we develop a clustering-based DAM.•Extensive evaluations on three video-based person re-id datasets have shown the effectiveness of our asymmetric modelling.

论文关键词：Person re-identification,Visual surveillance

论文评审过程：Received 14 August 2018, Revised 4 April 2019, Accepted 9 April 2019, Available online 4 May 2019, Version of Record 9 May 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.04.008