Human Activity Recognition in Ambient Assisted Living Environments using A Convex Optimization Problem

Human activity recognition (HAR) is of crucial importance in ambient assisted living (AAL) environments. In such environments, residents' data are collected via unobtrusive sensors to be further interpreted as human activities. Essential services are provided for the users according to the HAR results. In this paper, a novel scheme is represented for HAR, which is based on minimizing a convex objective function. For this purpose, given a sensor data stream, a belief vector (BV) is calculated for each sensor event. A BV is calculated by weighting the posterior probabilities of activities. It denotes the possibility of each activity, given a single sensor event. Then, the BVs are smoothed by minimizing a convex objective function. The objective function is formulated based on common assumptions in AAL environments. The final activity inference is based on the smoothed belief vectors. The proposed method is evaluated using a well-known and publicly available dataset. It is compared to three widely adopted HAR benchmarks, which are based on probabilistic learning graphical models. Simulations show that the proposed method outperforms the benchmarks, having an accuracy of 79.97% and an average F-measure of 78.2%.