I. Introduction
Automatic recognition of physical activities has emerged as a key research area in mobile and ubiquitous computing. Sensor-based activity recognition is important for assisted living, healthcare monitoring system, fall detection, and elderly support system. Vision based action recognition approaches are well-known for more than a decade but the progress is not enough. Plenty of methods are dealing different kinds of actions but little work has been done on elderly people's normal and critical activity detection like fall detection. A person may fall down and sustain injuries that limit mobility or encounter events that may lead to worsening health. The identification of inconsistent patterns, especially in daily activities monitoring systems, can be significantly valuable for caregivers to make decisions or diagnoses in emergency situations.