@phdthesis{def882d5-65d7-446c-8fbe-1721f6154beb,
  abstract     = {{Localization is fundamental in many applications, including autonomous driving, robotics, emergency services, and wireless communication systems. <br/>The achievable localization accuracy depends on the information provided by the measurements, which is shaped by sensor placement, spatial resolution, sampling rate, synchronization accuracy, calibration, noise levels, and environmental effects such as multipath propagation and occlusions. <br/>At the same time, practical localization systems must operate under constraints such as limited computational resources, energy consumption, and real-time processing requirements. As a result, localization should be viewed not only as an estimation problem but also as a broader system design problem that involves measurements from sensors, algorithm selection, and hardware implementation. By considering these aspects jointly, this thesis investigates how localization systems can be designed to achieve reliable and accurate performance while remaining suitable for practical real-time applications.<br/><br/>This thesis consists of introductory chapters that provide background on localization systems, followed by five papers published in or submitted to scientific conferences and journals. The first paper presents the Lund University Vision, Radio, and Audio (LuViRA) dataset, a public multimodal dataset developed through a collaboration between multiple departments and faculties at Lund University. It provides synchronized measurements from multiple sensor types for indoor robot localization and a common reference for developing, evaluating, and comparing methods across sensing modalities. The second and third papers build on LuViRA by investigating the accuracy, robustness, and practical limitations of different sensors, including a machine-learning-based 5G localization method using massive MIMO channel measurements. The fourth paper shifts the focus to outdoor radio-based localization and develops an adaptive attention-based model that selects between specialized machine learning models according to the propagation environment. Finally, the fifth paper extends this approach towards Transformer-based models and implements an FPGA accelerator to support low-latency and energy-efficient inference. Together, these contributions span measurements, algorithm development, environment-dependent adaptation, and hardware implementation, highlighting the importance of considering the complete localization system for practical real-time deployment.}},
  author       = {{Yaman, Ilayda}},
  isbn         = {{978-91-6858-033-3}},
  issn         = {{1654-790X}},
  language     = {{eng}},
  number       = {{195}},
  publisher    = {{Department of Measurement Technology and Industrial Electrical Engineering, Lund University}},
  school       = {{Lund University}},
  series       = {{Series of Licentiate and Doctoral Theses}},
  title        = {{Efficient Real-Time Localization Systems: Measurements, Algorithms, and Hardware}},
  url          = {{https://lup.lub.lu.se/search/files/260791771/Ilayda_Yaman_PhD_Thesis.pdf}},
  year         = {{2026}},
}