A Novel Implementation of an Autonomous Human Following Drone using Local Context

Abstract

This paper presents a novel implementation of an autonomous human following done using local context algorithm. The vision system consists of a drone equipped with a companion computer and camera. Through the vision system, drones can behave according to its desired application. In this study, the Pixhawk 2 Cube flight controller of the drone is connected to Odroid XU4, the companion computer. The OpenMV camera is connected to the Odroid XU4 and streams image bytes whereas the Odroid XU4 performs image processing to detect the human through local context. The Odroid XU4 controls the drone by sending commands to the flight controller based on detected objects. Drone following is implemented by detecting humans from the image stream provided by OpenMV and maintaining the detected human on the center of the image and within a specified distance through drone movements. The results show the single movements right, left, forward, and backward yielded low steady-state errors with settling time of about 3 to 4 seconds.