Third-person Immersion Vision-based Autonomous Target Tracking for Unmanned Aerial Vehicles

Abstract

In recent years, Unmanned Aerial Vehicles (UAVs) have been entering new application areas. The objective of this thesis is to implement autonomous target tracking for a quadrotor UAV through the use of vision-based technology. The purpose is to implement a third-person view experience, where the UAV hoovers behind and follows a targeted user, while sending a live video feed to a pair of virtual reality glasses, called Oculus Rift, worn by this user. Using visionbased tracking, one can minimise the overall equipment mounted on UAVs by utilizing the cameras that are already critical for the application area. Target tracking is achieved through the use of an image processing algorithm using Hough Circle Detection. A comprehensive analysis of this algorithm, also utilizing two different colour space filtering algorithms, is presented. Hardware choices made in building the quadcopter are also presented, supported by vibration test data showing the flight stability of it. This thesis proposes a candidate model for using autonomous UAVs, with image processing as main navigational tool, in a third-person view application. Due to noise in the transmission of video sent from the UAV to the ground station that transforms image data to maneuvering commands, target user recognition have not reached the preferred level of accuracy, which have negatively affected the general performance of the application. This issue is presented and discussed. Image processing is concluded to be a valid alternative as a way of implementing autonomous control of UAVs. Successful video transmission from the UAV to the Oculus Rift is demonstrated, as well as semi-successful autonomous control of the UAV.