@mastersthesis {pub4639,
	title = {SFA based Self Localization using Natural Landmarks},
	author = {Karam Sai Krishna Kaushik},
	year = {2020},
	month = {December},
	abstract = {The ability of a mobile robot to locate itself in an unknown environment is an essential criterion to implement any intelligent behavior. As cameras are becoming extremely
inexpensive these days, the vision-based localization is one of the most feasible options that one can choose to build a mobile robotic system. The fundamental aspect of the
robot{\textquoteright}s localization is the spatial representation which could be extracted using the unsupervised Slow Feature Analysis (SFA) algorithm. As a part of the previous work,
the robot was using the holistic view of an image captured via an omnidirectional camera to localize. To reduce the computational time and to remove the dependency
of learning orientation invariance during the training phase, the ongoing research had progress using a couple of techniques to extract smaller regions from an image to
train the SFA model. The state-of-the-art YOLO model has shown good results in object detection in the past decade. The current implementation uses a pre-trained
YOLO network model that learns scene specific objects in the garden to detect and extract the landmark regions. Another approach to extract the landmark regions is
by using uniquely encoded artificial markers such as ArUco markers. Thus, using the above techniques, a robot can localize itself faster than localizing it with the holistic
images, with improved accuracy of the model. We present the experimental results from outdoor environments and real-world recordings.},
	publisher = {UAS Frankfurt},
	booktitle = {UAS Frankfurt}
}