Classification of Activities of Daily Living Based on Depth Sequences and Audio
Chathuranga Siriwardhana, Danushka Madhuranga, Rivindu Madushan, Kutila Gunasekera · 2019
In this paper, we propose a non-intrusive daily activity recognition system to monitor elders who are living alone. Our contributions are to collect a dataset of activities of daily living consist of RGB, depth, silhouette, skeleton and audio data streams and to develop a system to recognize 20 different activities using depth image sequences and audio data. We have trained two separate Neural Networks to recognize activities from human silhouette features and Short Time Fourier Transform features extracted from depth images and audio data. Predictions from the two Neural Networks are fed into a fusion model and the final activity classification is done. The proposed activity recognition system can be used in assisted living systems to enhance the quality of the lifestyle of the elders who are living alone.