Robust human gesture recognition by leveraging multi-scale feature fusion

作者：

Highlights：

• Feature maps in the same size outputted by different layers are fused, which can effectively retain the key information on a small scale.

• The model utilizes feature transfer strategy for the region proposed network (RPN) within Faster-R-CNN model.

• This model combined several stages such as feature extraction, ROI region detection and gesture recognition into one step, which realized end-to-end training and testing mode.

摘要

•Feature maps in the same size outputted by different layers are fused, which can effectively retain the key information on a small scale.•The model utilizes feature transfer strategy for the region proposed network (RPN) within Faster-R-CNN model.•This model combined several stages such as feature extraction, ROI region detection and gesture recognition into one step, which realized end-to-end training and testing mode.

论文关键词：Gesture recognition,Faster R-CNN,Feature extraction,Human–computer interaction

论文评审过程：Received 9 July 2019, Revised 24 December 2019, Accepted 27 December 2019, Available online 31 December 2019, Version of Record 5 February 2020.

论文官网地址：https://doi.org/10.1016/j.image.2019.115768