Dilated Temporal Convolutional Neural Network Architecture with Independent Component Layer for Human Activity Recognition

A system using Convolutional Neural Network was developed for Human Activity Recognition (HAR) application. The proposed method uses the algorithm to detect the daily activities of people with cognitive disabilities. As an initial application, daily hygienic activities were monitored to assist people with cognitive disabilities who often need prompts to start and finish activities. The proposed system discussed in the paper uses Dilated Temporal Convolutional Neural Network Architecture with Independent Component Layers (DT-CNN-IC) to detect daily hygienic activities. The proposed system can be trained to obtain high average classification accuracy of 92.66 percent from only using a single channel accelerometer information. This method can reduce the necessary computational resources to train the proposed neural network allowing the algorithm to be implemented on low-cost platforms.