Activity Recognition using Deep Denoising Autoencoder

Existing feature extraction method for activity recognition is time consuming and laborious and prone to error. This paper proposes an unsupervised deep learning method for feature learning in activity recognition using tri-axial accelerometer. The proposed method extracts the relevant features automatically, eliminating the needs of feature extraction and selection stages. We evaluate and compared the proposed method with the conventional method in terms of recognition accuracy on a public dataset with wide range of activities. Results have shown that the proposed method achieved a better performance, improving the recognition accuracy by 0.03.