Research On Fall Detection Based On KINECT

In recent years,falling has become one of the most dangerous emergency situations for elderly people who live alone in their house,and in most cases,the seniors falling down cannot stand up and get immediate assistance,which could be critical for the elder.Computer vision-based systems provide a novel,low-cost and feasible solution for fall detection.And Kinect-based fall detection has become the main research direction.This paper proposes a fall detection method based on velocity and angle features.This method could be used to detect the falling of old people in the case of non occlusion.In order to detect falling,we extract two features:human head speed,and the angle between human major axis and the ground plane.Firstly,we track human head and calculate its speed,and decide whether it exceeds a certain threshold;and meanwhile the major axis of the people is estimated by approximating the coordinates of the shoulder center,spine,hip center,the angle between the major axis and the ground plane will be calculated and we also decide whether the angle exceeds a certain threshold.Only these two conditions are met at the same time,we will confirm that the fall has happened.According to the analysis of the characteristics of falling,we propose another fall detection algorithm.We find that in the process of falling,it will produce a tilt angle between the human body and the ground plane.When this angel is smaller,it shows that the human body is closer to the ground plane.In addition,falling is a transient process,and the duration of falling is very short.So we propose to utilize the angle changing rate to show fall.The bigger the angel changing rate is,the more likely the fall happens.So we will think the fall has happened when the angle changing rate is greater than a certain threshold.