An intelligent video surveillance system for fall and anesthesia detection for elderly and patients

Abstract

Abnormal activities detection is a major challenge to health care services, caregivers and families. Since some of these activities such as fall and anesthesia can be dangerous to health, we need to identify them with a good accuracy and speed. The detection of falling is categorized into three types consist of sensors and wearable devices, machine vision based and finally hybrid methods which are based on both sensors and machine vision approaches. We propose an automated vision based approach that detects moving objects in a given area using Gaussian Mixture Models (GMM) and filtering. Then, the system extracts some features from image of moving objects, processes changing them in consecutive key frames and triggers an alarm when a serious incident occurs to prevent possible future injuries. This real-time method synchronously detects fall and anesthesia using posture analysis with new fusion of features. Also, we propose an occlusion and overlapping handling mechanism in our system. According to experimental results, accuracy of fall detection and anesthesia identification are 93.59 and 86.11 percent respectively.