Dynamic fiducial markers for camera-based pose estimation

This dissertation introduces new techniques that increase the accuracy of camera-based pose estimation using fiducial markers. The problem of camera-based pose estimation involves finding a camera's pose relative to some coordinate system by detecting some known features in the environment; when the visual appearance of these features is known beforehand, they are called fiducials. The visual-based pose estimation process is highly complex since the estimated pose accuracy depends on many interconnected factors that have to be considered simultaneously; this thesis aims to identify the most influential factors and proposes solutions that mitigate the effect of the sources of error, hence increasing the estimated pose's accuracy and robustness. We base our solutions on exploiting an interaction between the camera and what the camera is measuring; this, in essence, means that the features change and adapt to better suit the measurement by either moving in space to better locations or changing their shape dynamically