Geometric Kinect Joints Computing for Human Fall Recognition

Abstract

This paper proposes an computing analysis on human geometric shape features to detect a fall behavior. The system mainly computes the changes on human orientation (torso angle) and centroid height via the human skeleton joints extracted by Kinect sensor. The system computes and tracks the spatial changes of these human orientation and centroid height and distinguishs a fall behavior among other daily activities by using a thresholding algorithm. The main objective of this computation is to minize the computational time and to increase the true alarms in developing a fall detection. The system works the feature extraction on our collected fall detection dataset containing the fall data along with daily activities such as sitting down, lying, combing. Standing, etc., are collected by Microsoft Kinect sensor.