Smartphone Based Human Activity Recognition Using 1D Lightweight Convolutional Neural Network

Smartphones are an obvious platform for the deployment of the Human Activity Recognition(HAR) system. But, they are limited in terms of processing power, energy and storage space. Therefore, there is a need to make lightweight deep learning models that can be run within these constraints. In this paper, we propose a one-dimensional lightweight Convolutional Neural Network(CNN) that can be operated on smartphones. In the proposed one-dimensional lightweight CNN model, pruning and quantization are used to compress CNN model size without significant accuracy losses. The experimental result shows that the proposed CNN model was proven to be successful accuracy while maintaining their performance even after quantization and pruning.