A Vision-based Approach for Human Fall Detection in Indoor Environment

Abstract

The population of elderly citizens in the world has been increased in every year. With the growing of a population most of the elderly citizens become to live independently and not have all day daycare of their loved ones. Therefore, home safety for the elderly people is an important issue in the current society. Falls are one of the major risks for elderly people living alone at home. In most cases they cannot help themselves and asses rst aid after a fall. Providing immediate help just after a fall reduces the risk of the fall. Therefore, e cient and accurate detection of a falls is critical, and prompt noti cation to the outside is essential within a short time period. As a result, many fall detection systems have been developed. The existing solutions consist of wearable, ambience and vision based approaches. The wearable devices are popular and these devices must be worn on di erent locations of the body. However, these devices can be rather expensive, uncomfortable, and also most of the people forget to wear them. This research proposed a novel person adaptive vision based fall detection approach which can work real-time, on day and night. The proposed fall detection approach consists of two phases which semantically analyze the fall characteristics and detects falls. An Approach used depth sensor and integrates 3D skeleton motion and 3D oor plane. The method requires joint to the oor and joint vertical velocity features of the head, spine and hip center joints for a fall detection. The fall validation is done by novel thresholds based algorithm and threshold values changed with the person`s height. The proposed approach evaluated with 15 volunteers with standard evaluation criteria.The approach gave 97.77% of speci city , 89.77% of sensitivity and 93.82% of accuracy values. Issues and proposed improvements to the system are also discussed in this work.