3D-based Deep Convolutional Neural Network for action recognition with depth sequences

作者:

Highlights:

• Studies machine learning-based action recognition from depth sequences

• Directly learns spatio-temporal features without using hand-crafted features

• Our feature representation is time-invariant and viewpoint-invariant.

• Outperforms or achieves comparable results to the state-of-the-art methods

• The trained model and learned features can be transferred to different datasets.

摘要

•Studies machine learning-based action recognition from depth sequences•Directly learns spatio-temporal features without using hand-crafted features•Our feature representation is time-invariant and viewpoint-invariant.•Outperforms or achieves comparable results to the state-of-the-art methods•The trained model and learned features can be transferred to different datasets.

论文关键词:Action recognition,Deep learning,Convolutional neural network,Depth sequences,3D convolution

论文评审过程:Received 28 June 2015, Revised 4 February 2016, Accepted 7 April 2016, Available online 16 April 2016, Version of Record 10 November 2016.

论文官网地址:https://doi.org/10.1016/j.imavis.2016.04.004