EAR: An Energy Efficient Human Activity Recognition from Wearable Devices

Human Activity Recognition (HAR) plays a crucial role in wireless body area networks. This article presents an energy-efficient model based on a combination of pruned Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU). This model will be relevant for lightweight gadgets due to low memory requirements and reduced model complexity. The proposed model effectively captures the spatial and temporal features from sensor data and leverages the combination of Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU). We applied the pruning technique to the CNN+GRU layers to improve energy efficiency by reducing the model complexity and computation. We have evaluated the proposed EAR model on a publicly available human activity dataset collected from wearable sensors. We enhanced the efficiency of the aforementioned combined approach by implementing pruning techniques on CNN. We demonstrate that our lightweight human activity recognition method achieves an accuracy of 99.49 % for UCI-HAR datasets. The proposed method is useful for devices with limited memory capacity.