Elderly fall detection with depth camera

The elderly people are frequently exposed to falls due to biological and physiological changes in the human body in older ages. This situation causes injuries in elderly people. Treatment processes are prolonged and adversely affects the life quality of old people. Injuries also cause substantial increase in the country's health expenditures. Detection of the elderly falls is of great importance in order to solve these problems. For this purpose, in thesis study three different methods are proposed in order to detect fall in elderly people with depth camera and a data set is formed. 1. Silhouette Orientation Volumes (SOV) feature extraction method is proposed in order to define motion and stance based shape features which enable fall detection from depth images. In applications carried out with SDU-Fall data set, 91.89% success rate in fall detection and 89.63% success rate in motion detection are obtained. 2. For coding the features Fisher Vector (FV) coding method is suggested in order to develop the distinguishability of fall from daily activities such as walking, sitting and lying in depth images. In applications performed with the proposed method and SDU-Fall data set, 88.83% success rate in fall detection is achieved. 3. Feature extraction algorithm based on reducing three dimensional (3D) skeleton joint data to two separate two dimensional (2D) skeleton joint data is proposed. 97.08% success in fall detection is obtained in applications with FUKinect-Fall data set. 4. FUKinect-Fall dataset which contains 1008 depth and skeleton joint data that simulate walking, bending, sitting, squatting, lying and falling actions of 21 subjects by using Kinect camera is formed. By comparing the results obtained from studies with the results of similar studies in the literature, the success of the proposed methods is presented.