Multi-Channel Deep Networks on Sequence Data for Multi-Action Recognition
Qianqian Qi, Siwei Zhong, Ye Wang, Jun Bo Long · 2020
Recently, Human Activity Recognition (HAR) using deep neural network has become a hot topic in human-computer interaction. Machine can effectively identify human naturalistic activities by learning from a large collection of sensor data. Distinguishing the same activity of different human is not only an interesting research problem, but also has real applications in private identity authentication. Based on the success of convolutional networks for sequence raw data and residual network to achieve the level of aesthetic representation of the automatic learning, we propose a novel model, named Deep Convolutional and Residual Networks (ConvResNets), for identifying authentication on wearable data, in which: (1)The raw input features can be replaced by high-level features representation; (2)ConvResNets does not require expert-knowledge to extract activity data features; (3)ConvResNets can easily capture features distribution of private activities. We evaluate our model on two public datasets. Our results show that our model is effective to recognize identity authentication via wearable datasets. We discuss the influence of networks parameters on performance to provide insights about its optimization.