Visual monitoring systems for elderly and newborns for healthcare services

Abstract

Whereas in the hospital environment image analysis is frequently applied for diagnostics, video surveillance involving analysis is not yet widespread, while it can increase both healthcare and nurse efficiency. This paper explores the possibilities of automatic video analysis in this environment for two applications: detection of falling elderly and discomfort detection of newborns. For fall detection, two algorithms are proposed that aim at detecting fall incidents during different stages of activity (walking, sitting, etc.). The first algorithm is based on the standard case with two cameras in perpendicular orientation. In this case, a high performance is obtained (on our test set, we achieve 100% in unoccluded situations). The second algorithm operates in a more challenging case with one camera only. This algorithm can approach a high accuracy under certain conditions, but is less robust. In contrast to systems described in literature, both proposals are able to handle the use of walking tools and are tested at night-time, using Infra- Red lighting. For discomfort detection, an algorithm is proposed that analyzes the facial expression, based upon measurements in the eye, eyebrow and mouth regions. The algorithm employs a hierarchical classifier to discriminate between the states sleep, awake and cry. For the available test sequences, the algorithm is able to operate in a wide range of lighting conditions and viewpoints. While both systems promise to become helpful in healthcare, validation in the real hospital environment is an upcoming stage in the actual project. It was found for the elderly study that clothing color in the hospital environment can be of help in preserving detection robustness during various activities.With respect to discomfort detection, the use of markers on medical equipment and baby-attached sensors would greatly simplify the separation of those devices from the facial details.

Date of Award	31 Oct 2008
Original language	English
Supervisor	Peter H.N. de With (Supervisor 1)