MSST-ResNet: Deep multi-scale spatiotemporal features for robust visual object tracking

作者：

Highlights：

• Firstly, we set up a residual learning framework to combine multiple different residual units by adding skip connection and multi-scale feature.

• Secondly, we present a novel tracker based on deep multi-scale spatiotemporal features model. Our tracker can successfully locate the target object in the consecutive video frames.

• Finally, the experimental results show that our tracker increases the tracking performance substantially.

摘要

•Firstly, we set up a residual learning framework to combine multiple different residual units by adding skip connection and multi-scale feature.•Secondly, we present a novel tracker based on deep multi-scale spatiotemporal features model. Our tracker can successfully locate the target object in the consecutive video frames.•Finally, the experimental results show that our tracker increases the tracking performance substantially.

论文关键词：Visual object tracking,Residual network,Kernelized correlation filter,Spatiotemporal features,Multi-scale features

论文评审过程：Received 2 November 2017, Revised 28 October 2018, Accepted 30 October 2018, Available online 9 November 2018, Version of Record 19 December 2018.

论文官网地址：https://doi.org/10.1016/j.knosys.2018.10.044