The aging of the baby boomers, the rise of the Internet of Things (IoT), and skyrocketing medical costs have all converged to expand the market for remote health monitoring. There’s a tremendous push to enhance elder care especially, and eventually all disability and rehabilitative care, through new technology.

The authors of this study begin by documenting both the need for, and the history of, research into automated, ambient behavior monitoring. With ambient monitoring, sensors are unobtrusively located in the environment, as opposed to being placed on the body or requiring interactivity. Anyone interested in automated behavior analysis would value the paper’s references to previous research alone, notwithstanding the actual content of the paper.

After this summary of prior work, the authors present their contributions to the field. First, they use Bayesian estimators to improve the statistical recognition of anomalies. Then, they combine three types of sensor usage to improve anomaly recognition: sensor activation likelihood (SAL), to monitor circadian rhythms for long-term status monitoring; sensor sequence likelihood (SSL), to recognize confusional states; and sensor event duration likelihood (SDL), to recognize physical conditions such as falls or unresponsive states.

The remainder of the paper presents data collected on three test subjects along with analysis of the system’s response to that data. The authors’ approach was shown to yield improved recognition of anomalies, with the caveat that the severity of the anomaly can’t be determined. That is, while the system can reliably signal a change in behavior, it has no semantics for reporting the medical significance of that change. More research is called for in order to continue improving and tuning the model.