Dynamic background subtraction for fall detection system using a 2D camera

Falls are the number one cause of fractures, trauma, loss of independence and death among older adults. Each year more than 1.6 million older U.S. adults go to emergency departments due to falls [1]. It is necessary to provide mechanisms to detect people´s falls in order to improve their lives. This paper proposes a fall detection system that utilizes a 2D camera. It employs an open source library (OpenCV) and computer vision techniques in order to detect falls. Background subtraction (dynamic) and movement detection are used to detect falls. Flags and timers are employed to improve the system. Flags are used to avoid false positives when the starting position is the floor. Timers determine the minimum amount of time a person should lie on the floor in order to detect a fall the system was implemented and evaluated obtaining an accuracy of 85.37% of fall detection.