DNN and CNN Approach for Human Activity Recognition

Abstract

One of the common causes of low back pain is postural stress. When sitting or walking, poor posture may result in spinal dysfunction. Increased pressure on the spine can cause tension and spasms in the lumbar muscles and cause low back pain. Monitoring of daily activities becomes more important, especially to help sick and elderly people. Recognition of unstructured daily activities is a more difficult and important task. In this study, we use Deep Neural Networks (DNN) and Convolutional Neural Networks (CNN) to study spinal movement and postural stress through two sensors connected to the pelvis and spine of a healthy subject. Body kinematics data consist of four categories: standing, sitting, walking and other activities. We compared the accuracy of DNN and CNN methods for the identification and labeling of daily activities. We observed the results of deep learning methods with different hyperparameter values and obtained the optimum values.