Efficient large-scale action recognition in videos using extreme learning machines

作者：

Highlights：

• We describe a novel approach for large-scale action recognition from videos in a realistic setting.

• We represent each video by Fisher vector encoding computed on improved trajectory features.

• We use extreme learning machines for fast and accurate classification.

• We report comparative results where we show that the proposed approach outperforms the baseline approaches, and reaches accuracy close to state of the art without using deep neural networks.

摘要

•We describe a novel approach for large-scale action recognition from videos in a realistic setting.•We represent each video by Fisher vector encoding computed on improved trajectory features.•We use extreme learning machines for fast and accurate classification.•We report comparative results where we show that the proposed approach outperforms the baseline approaches, and reaches accuracy close to state of the art without using deep neural networks.

论文关键词：Action recognition,Extreme learning machine,Fisher vector,Multimedia mining

论文评审过程：Available online 19 June 2015, Version of Record 25 July 2015.

论文官网地址：https://doi.org/10.1016/j.eswa.2015.06.013