Smartphone-based Human Activity Recognition in buildings using Locality-constrained Linear Coding

Abstract

Smartphone-based Human Activity Recognition (HAR) is an active research topic in smart homes and health care. Smartphones are often carried by users in a non-intrusive way, and the sensing platform in them could provide much information about the users. By leveraging the built-in accelerometer and gyroscope in smartphones, we can design a system to distinguish the user's simple behaviors. Most of the current work did not consider feature selection, but directly fed the statistical features from both the time and frequency domains into machine learning algorithms. In this work, we proposed an approach for HAR by using Locality-constrained Linear Coding (LLC) as a feature selector to improve the performance of HAR systems. After feature selection, standard typical classifiers of Support Vector Machine, K-Nearest-Neighbor, Kernel-Extreme Learning Machine and Sparse Representation Classifier can then be applied to learn the distinctiveness of the selected features. Experiment results showed that the LLC approach achieved an average accuracy of about 90% due to a better selected dictionary for feature representation. Activities of simple actions that occupants usually perform in buildings include walking, walking upstairs and downstairs, running and static. This work on occupant behaviors could be used in energy-efficient building and health care applications.