A Parallel Network used to Human Activity Recognition
Changhao Ni, Shengxiao Guan · 2021
This paper proposes a novel module AIC for HAR task, which can achieve a good accuracy with few parameters relatively. Instead of blindly improving the model size, we utilize the parallel convolutional network to fully extract features of different dimensions. Extensive experiments on three publicly available HAR datasets are conducted to validate the generalization of our proposed module. The specific results are presented in Experiment Results and Analysis, which shows that the module consumes less time and parameters and is promising to recognize human activity in Mobile devices.