Accuracy Evaluation of Methods for Pose Estimation from Fiducial Markers
Budak, Ugurcan (2021)
Budak, Ugurcan
2021
Automaatiotekniikan DI-ohjelma - Master's Programme in Automation Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2021-12-10
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202112088996
https://urn.fi/URN:NBN:fi:tuni-202112088996
Tiivistelmä
Estimating the pose from fiducial markers is a widely researched topic with practical importance for computer vision, robotics and photogrammetry. The real-world use cases can be synthetically replicated and it is easy to model errors in this way. Despite the simplicity of the pose estimation process, various parameters in the process can affect the accuracy. The frequently evaluated error in pose estimation problem is an error type that arises during the detection of the markers by an image sensor. Nevertheless, in real-world use cases, there are several other types of errors that are expected to appear. As markers are the physical entities in the real-world use cases, the number and manufacturing process of the markers, settlement of them into the world scene, the size of the area occupied by the markers typically result in errors in the pose estimation.
In this thesis, the aforementioned error types are modelled based on the real-world use cases. Using perspective projection, with a calibrated camera, the errors are quantified for four different pose estimation algorithms. Even though these four non-iterative algorithms focus on the pose estimation problem to find the global minima of their cost functions, their behaviours considerably differ from each other. The empirical results are elaborated and the algorithms are compared with each other. The results reveal that the modelled error types significantly affect the rotational and translational accuracy of pose estimation. Considering the empirical results, while the errors resulting from detection, manufacturing and settlement of the markers leads to a decrease in the accuracy, an increase in the number of markers and the constellation size leads to an increase in the accuracy of pose estimation.
In this thesis, the aforementioned error types are modelled based on the real-world use cases. Using perspective projection, with a calibrated camera, the errors are quantified for four different pose estimation algorithms. Even though these four non-iterative algorithms focus on the pose estimation problem to find the global minima of their cost functions, their behaviours considerably differ from each other. The empirical results are elaborated and the algorithms are compared with each other. The results reveal that the modelled error types significantly affect the rotational and translational accuracy of pose estimation. Considering the empirical results, while the errors resulting from detection, manufacturing and settlement of the markers leads to a decrease in the accuracy, an increase in the number of markers and the constellation size leads to an increase in the accuracy of pose estimation.