An activity of daily living primitive-based recognition framework for smart homes with discrete sensor data

The proven approach successfully recognizes the activity of daily living is a classifier training on feature vectors created from streamed sensor data. However, there is still room to improve feature extraction techniques in that the activity of daily living data are often nominal or ordinal. The ordinal data can be likely less discriminative due to the great uncertainty in level of measurement. This article provides a framework with novel activity of daily living primitive that introduces an enhanced feature selector with linear time complexity. The extension to traditional approaches is that the present framework considers the following: (1) defining activity of daily living primitives and constructing a primitive vocabulary, (2) reducing data when representing raw activity data, and (3) selecting an appropriate primitive set for each testing activity. The empirical results reveal that a pre-trained portable primitive vocabulary not only outperforms the existing baseline frameworks but also greatly facilitates the deployment and management of activity recognizers.