Comparison of radar sensors for indoor human localization

In the last decades, the growth of the GPS usage in many fields for outdoor localization has led to a higher demand of systems for indoor human localization for many fields of application, like home security or safety and healthcare. Creating an indoor localization system presents numerous challenges due to several characteristics that vary in every considered environment, such as room dimension, roof height, or furniture type and location. The recent improvements in the computer vision field given by machine learning algorithms can provide a possible solution to this problem, by using optical cameras that are able to follow human movements. Still, such systems have many problems: privacy, cost, clear line-of-sight need. The aim of this thesis work is to investigate the indoor localization techniques using two radar sensors with different frequency bands, in order to obtain a human tracking system with the lowest possible error. The two devices utilized are commercial evaluation kits, complete with the microcontroller needed to process the measured signals, in order to provide to the host information in terms of human target(s) detected in the area covered by the sensor and some parameters associated to such target(s). The first device is the DEMO Position2GO provided by Infineon, based on the 24 GHz transceiver chipset (BGT24MTR12) and the ARM® Cortex™-M4 XMC4700 microcontroller, while the second radar is the IWR6843ISK by Texas Instruments, based on the 60-64 GHz transceiver chip (IWR6843) and the processor sub-system, composed by the Digital Signal Processor (DSP) sub-system (C674X DSP) and main processor system (ARM® Cortex™-R4F microcontroller). The main challenge of the work was to find the best sensor configurations and to improve the obtained measurements via post-processing to reduce the errors for the three tested environments: a) Laboratory, a small room where the measurement area is surrounded by furniture; b) Corridor, an empty large indoor area inside the university building; c) Terrace, an empty outdoor terrace at the top of the university building. Ideally, a good sensor should provide measurements that reflect closely the real movement of the target person, have a high self-consistency, e.g., the device should report the same coordinates each time the person is in the same position at different times, and the measurements should accurately reflect the dynamics of the movement of a human body. The obtained results are evaluated with specific metrics and allow us to evaluate the best sensor configurations to obtain a good accuracy between real and measured positions and a good sensor self-consistency. The quality of the sensor measurements is improved in this work via error minimization performed on the quality metrics. These findings can be further improved using more advanced post-processing algorithms, including for instance machine learning algorithms. In addition, the relatively low price, the multiple-choice of commercial solutions, and the relative ease of implementation of human localization and tracking systems based on such devices in various existing indoor environments demonstrate that the radar technology can be a good candidate for Indoor human localization systems.