Daily Human Activity Recognition Using Depth Silhouettes and Transformation for Smart Home

We present a human activity recognition (HAR) system for smart homes utilizing depth silhouettes and

$\mathcal{R}$

transformation. Previously,

$\mathcal{R}$

transformation has been applied only on binary silhouettes which provide only the shape information of human activities. In this work, we utilize

$\mathcal{R}$

transformation on depth silhouettes such that the depth information of human body parts can be used in HAR in addition to the shape information. In

$\mathcal{R}$

transformation, 2D directional projection maps are computed through Radon transform, and then 1D feature profiles, that are translation and scaling invariant, are computed through

$\mathcal{R}$

transform. Then, we apply Principle Component Analysis and Linear Discriminant Analysis to extract prominent activity features. Finally, Hidden Markov Models are used to train and recognize daily home activities. Our results show the mean recognition rate of 96.55% over ten typical home activities whereas the same system utilizing binary silhouettes achieves only 85.75%. Our system should be useful as a smart HAR system for smart homes.