Pose error estimation of a quadcopter in the outdoors

Date
2016-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: The quadcopter industry is a fast-growing and maturing industry which pro- duces `dumb' (or manually controlled) unmanned aerial vehicles (UAVs). They are also commonly equipped with controllers which make them able to au- tonomously carry out a ight mission without a human pilot. Industry is becoming increasingly interested in integrating quadcopters into their respec- tive workforces in an attempt to automate some processes. However, national governments are worried that if left unregulated, UAV quadcopters may pose a safety and security threat to society, particularly if they are to work au- tonomously in the outdoors. There are a number of improvements that can be made to increase the safety of quadcopters. This research project investigates how well quadcopters can estimate their position and orientation, or pose. If this is known, it can be integrated into a control model as an error term to improve the performance and safety rating of a quadcopter. Indoor measurement tools cannot be used, since the quadcopters of interest rely on GPS data. Therefore, a computer vision-based pose measurement system (CVS) was investigated, designed and implemented to measure the pose of a quadcopter in the outdoors. The system's measurements were compared to a quadcopter's pose estimates recorded during an outdoor test ight. The results show that the CVS's measurements are more accurate than the quadcopter's in all the dimensions except for the yaw. It was found that the quadcopter's position estimation error is approximately 150mm for x, y and z, and 3.27° and 1.9° for the roll and pitch dimensions. These results can be used and integrated into a quadcopter's control model.
AFRIKAANSE OPSOMMING: Die vierbeen onbemandevliegtuigbedryf is 'n snelgroeiende industrie wat vlieg- tuie produseer wat met die hand of heeltemal outonoom, sonder 'n menslike vlieënier se inset, 'n vlugopdrag kan voltooi. Industrie stel toenemend belang daarin om sulke onbemande vliegtuie in hul werksmag te integreer om prosesse tot 'n mate to outomeer. Wêreldsregerings is egter bekommerd dat indien die onbemandeveliegtuigbedryf ongereguleerd gelaat word, sulke vliegtuie 'n ge- vaar sal inhou vir die samelweing, veral as hulle gelaat word om outonoom in die buitelug te werk. Daar is verskeie verbeterings wat gemaak kan word om die tegnologie se veiligheid te verhoog. Hierdie navorsingsprojek stel ondersoek in om vas te stel hoe akkuraat 'n onbemande vliegtuig sy posisie en oriëntasie kan afskat. Indien hierdie waardes bekend is, kan dit geïntegreer word in 'n beheermodel om so 'n vliegtuig se werksverrigting en veiligheid te verhoog. Binnemuurse meetinstrumente kan nie gebruik word hier nie, aangesien die onbemande vliegtuie van hierdie projek staatmaak op hul GPS lesings. Dus, in hierdie projek is daar 'n rekenaarvisiestelsel ontwerp, getoets en geïmplimen- teer om 'n vierbeen helikopter se posisie af te skat in die buitelug. Die vierbeen helikopter se afskattingsakkuraatheid was dan gevind deur sy afskatting te ver- gelyk met dié van die rekenaarvisiestelsel in 'n buitemuurse vlugtoets. The resultate toon dat die rekenaarvisiestelsel se metings meer akkuraat is as die vierbeen helikopter s'n vir alle meetdimensies, buiten die afwykingshoek. Dit was gevind dat die helikopter se posisieafksattingsfout ongeveer 150mm is in die x, y en z dimensies, terwyl die hoekafskattingsfout 3.27° en 1.9° is vir die rol- en hellingshoeke. Hierdie resultate kan geïmplimenteer word in 'n vierbeen helikopter se beheermodel.
Description
Thesis (MEng)--Stellenbosch of University, 2016.
Keywords
Computer vision, Neural networks, Unmanned aerial vehicle, Measurement accuracy, UCTD
Citation