A HUMAN FOLLOWING ROBOT FOR FALL DETECTION

Abstract

This thesis presents work on a human following robot for detecting falls in the home of the elderly. The goal is to have a robot that can detect a human, follow the human in a cluttered space, and determine when the human falls. A Raspberry Pi based Robot known as Fall Detection Robot (FADER) that had been developed in the Real-time and Embedded Control, Computing, and Communication (REC3) Lab at Middle Tennessee State University is used, and a number of adjustments are made to its design including adding a Pi Camera and an Arduino microcontroller board. Computer vision deep learning-based object detection is used as the means of detecting the human, and linear regression and threshold-based algorithms are used to estimate the distance to the human, navigate and to determine falls. The advantages of using FADER for fall detection include its being mobile, the user not being required to be involved for the technology to work, and its being non-invasive with respect to the user’s body. Furthermore, FADER is low-cost and easily manufacturable. Results show that the modified FADER functions with a high precision of 100% but low sensitivity of 42%.