Estimating human activity patterns in dynamic environments based on smart, wearable sensors : a feasibility study.

Abstract

The demands for understanding human activities have steadily grown in recent years in the health-care domain, especially in elder care support, rehabilitation assistance, diabetes, cognitive disorders, assisting living and wellness management. A significant amount of resources can be saved if sensors can help caretakers record and monitor elders or patients continuously and report automatically when any abnormal behavior is detected. The recognition of various activities of daily living (ADLs) can reveal valuable information about a person’s activity patterns. Many studies have successfully identified activities using wearable sensors with very low error rate, but the majority of the previous works are done in very constrained settings. Readings from multiple body-attached sensors achieve low error-rate, but the complicated setting is not feasible in practice. On the other hand, smartphones have been accepted from the research community as a powerful solution for sensing applications due to the increasing number of smartphone users and due to the vast capabilities of modern smartphones. This project uses low-cost and commercially available smartphones as sensors to identify human activities. The growing popularity and computational power of smartphone make it an ideal candidate for non-intrusive body-attached sensors. Unlike many previous reported works, we relaxed the constraints of attaching sensors to fixed body positions with fixed device orientation. In our design, the phone can be placed at any position around waist such as jacket pocket and pants pocket, with arbitrary orientation. In this work a feasibility study has been contacted to investigate whether a smartphone based recognition system can be used for estimating activity patterns in dynamic environments. To this end, different combinations of computational approaches have been taken. The computational pipeline was applied on separate activities of daily living and on complete sequences of activities, which describe a common scenario of daily living. The results showed that, using a 1 second-window with 80% overlap, the suggested feature sets and the k-NN classifier, the ADLs and the scenarios can be recognized with accuracy of 99% and 96 % respectively, when the 10-fold cross-validation evaluation method is applied. A further investigation using the aforementioned combinations and the evaluation method of Leave-One-subject-Out for the recognition of scenarios achieved accuracy of 79%.

-