Healthcare video surveillance: toward automated monitoring systems

加拿大正在经历人口的重大变化，人口老龄化非常迅速。因此，必须在医疗保健领域进行创新，以确定适当的方法，利用技术找到解决老年人所遇到问题的巧妙办法。因此，本研究项目的目的是开发和适应老年人的自动视频监控方法，使他们能够保持自主性，享受更多的日常生活。该研究分为三个不同的子项目，所有子项目都需要使用单个或多个摄像机进行自动视频监控。在第一个子项目中，我们提出检测老年人意外跌倒，这是独居老年人最大的危险因素之一。我们的方法是基于人体形状变化和身体的三维运动的组合，特别是头部。在同样的背景下，我们打算开发一种方法来测量老年人在他/她的公寓里的步态，以评估可能的退化或变化，以防止跌倒和/或诊断其他问题（活动能力降低，平衡能力下降，身体僵硬度增加，步伐缩短，步伐减慢，四肢运动减少等）。该方法采用简化的三维关节骨骼模型对步态进行分析。最后一项研究包括用于监测药物摄入的计算机视觉系统，这是老年人问题的另一个主要来源。该系统的核心是脸、手和药（瓶）的定位和跟踪，然后是多层次场景模型的活动识别。我们认为，这种视频系统将通过克服许多限制，对用于远程监控的其他类型的传感器进行有益的补充。

Canada is experiencing significant changes in its population, which is aging very quickly. It is therefore essential to innovate in the area of healthcare to identify appropriate ways that technologies can be used to find ingenious solutions to the problems experienced by the elderly. The purpose of this research project is therefore to develop and adapt automatic video surveillance methods for elderly so that they can preserve their autonomy and enjoy more their daily lives. The research is divided into three different subprojects all requiring automatic video surveillance with single or many cameras. In the first subproject, we propose to detect accidental falls of elderly people, which are one of the greatest risk factor for seniors living alone. Our approach is based on a combination of human shape variation and 3D motion of the body in general and the head in particular. In the same context we intend to develop a method to measure the gait of the elderly in his/her apartment for assessing possible degradation or changes to prevent falls and/or diagnose other problems (reduced mobility, decreased balance, increased body stiffness, shorter step, slower pace, reduced motion of the limbs etc.). The method uses a simplified 3D articulated skeleton model to analyze the gait. The last study consists in a computer vision system for monitoring medication intake, another major source of problems for elderly persons. At the core of this system are face, hands and pills (bottles) localization and tracking followed by activity recognition with multi-level scenario models. We believe that such video systems will advantageously complement other types of sensors used for telemonitoring by overcoming many of their limitations.