A Combined Stochastic Memory Model for ADL Detection

Many HAR (Human Activity Recognition) systems are able to detect sequential executed ADL (Activity of Daily Living). While a person is capable of doing two things in parallel or to pause one ADL and finishing it later a HAR system (HARS) must be capable to remember ADL states and decide which ADL is completed and which might be continued after the current ADL. We address this case by combining a stochastic Markov Model and a psychological memory function to detect parallel and nested ADL. For the evaluation, we use an input dataset and benchmark for comparison, which is publicly available [1]. Our approach outperforms the leading HARS for this benchmark by 2% points while using a more cost effective installation environment. Furthermore we address an unsupervised learning method to train the HARS and explain the algorithm of parallel ADL detection in detail.