Vision Based Navigation of an Autonomous Quadrocopter

OData support
Supervisor:
Dr. Stumpf Péter Pál
Department of Automation and Applied Informatics

The unmanned aerial vehicles have been gaining increasing attention during the past

decade. Besides their original field of application, the military, their commercial and scientific

use is now expanding as well. Due to their fairly low retail prices, they have become

accessible to a wide range of people. The most common commercial use of drones is aerial

photography, therefore most of them are equipped with at least one camera. Some of the

manufacturers of these remotely-controlled drones also provide application programming

interfaces (API) for software developers to create augmented reality games. The presence

of the API and the cheap hardware also make these drones a suitable platform for research

related to image processing and autonomous aerial vehicles.

The aim of this project is to develop a PC application using the API of a Parrot AR.Drone

2.0 quadcopter that enables the drone to detect, and follow a pedestrian autonomously.

In order to achieve this, various tasks had to be solved, which are presented in this MSc.

thesis. First, the structure of the drone is introduced. After this the image processing

problems related to both camera calibration and pedestrian detection are described. The

document presents and compares four types of human detection algorithms. Two are based

on colour-based segmentation, where the constraint is that the pedestrian has to wear an

unicolour T-shirt. After this, the use of a Histogram of Oriented Gradients descriptor-based

detection method is described. In this case no constraint is needed regarding the

appearance of the pedestrian. As a fourth type, the latter is combined with optical flow

calculations to present an algorithm more suitable for pedestrian tracking and following.

The image processing topics are followed up by the description of the drone's motion control

system. This includes the design of various controllers along with plant identifications, and

the calculation of the reference signals based on the camera image. The developed PC

application is also introduced briefly. The application was written in C#, and it's also

using third party libraries, such as the Emgu CV, a .NET wrapper for the widely-known

OpenCV image processing library. The end of the document also presents images and

videos showing the autonomous operation.

Downloads

Please sign in to download the files of this thesis.