A Deep Learning based Human Activity Recognition Model using Long Short Term Memory Networks

• Published in: 2022 International Conference on Sustainable Computing and Data Communication Systems (ICSCDS) pp. 1371 - 1376
• Main Authors: Durga, K. Mohana Lakshmi, Jyotsna, P., Kumar, G. Kranthi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.04.2022
• Subjects: accelerometer; Accelerometers; Activity recognition; Artificial Neural Networks (ANN); Costs; Deep learning; gyroscope; Human Activity Recognition (HAR); Legged locomotion; Long Short-Term Memory (LSTM); Training; Wearable computers
• DOI: 10.1109/ICSCDS53736.2022.9760794

The recent development of machines which shows different intellectual characteristics are being created by making changes in system hardware and software. In this modern period different hardware tools, smart wearables, machine learning models, and deep learning models are being applied in Human Activity Recognition (HAR) applications. The major significance of the problem is to develop a HAR model with high accuracy rate using lower cost sensors. To reach this goal, the sensor-based data are collected from two low cost sensors namely accelerometer and gyroscope sensors, which are used along with the construction of an Artificial Neural Networks (ANN) based deep learning model. Long Short-Term Memory (LSTMs) Networks are used to make the model learn and recognize the type of activities that the user is performing. The model is trained using UCI-HAR [9] dataset. In the HAR model, a person's activity is identified based on sensor readings. Here, six distinct activities such as sitting, walking, laying, standing, up-stairs and down-stairs are identified and recognized. Knowing the activities performed by users it can help in having instant interaction with them.