Indoor localization using fingerprinting of ultra-wideband measurements

Transcription

1 Contents Indoor localization using fingerprinting of ultra-wideband measurements... 2 Model based improvement of UWB-based indoor localization... 3 Robust 2d and 3d Helmert transformation with uncertainties in both coordinate frames... 4 Approaches and tools for geodetic research data management: the example of terrestrial laser scanning... 5 Modelling and detecting mixed pixels in laser scanning... 6

2 Lorenz Schmid, Dr. David Salido Monzú, Indoor localization using fingerprinting of ultra-wideband measurements Ultra-wideband (UWB) ranging may be a highly attractive techlogical basis to develop indoor localization solutions. The time resolution of these radiofrequency signals enables precise delay-based distance measurements to multiple des, which can in turn be used for localization. However, the accuracy of the resulting position estimates is significantly degraded in case of multipath interference due to signal reflections which are typical in complex indoor environments with many rooms and objects. While introducing significant ranging errors, the multipath effects exhibit high spatial repeatability when the main geometrical characteristics of the environment remain unchanged. This feature, together with the penetration properties of UWB signals enabling through-wall measurements, suggests that a fingerprinting approach can potentially be applied to UWB measurements for multi-room localization in complex indoor environment. The goal of this thesis is to investigate the feasibility of such a solution and assess how it compares to standard ranging-based UWB positioning in terms of accuracy, coverage, cost of setting up and operation, and computational cost. The analysis shall take both range and signal strength into account as potentially useful signals for fingerprinting. The investigation shall include representative test measurements based on the deployment of a multi-room UWB network within the HIL building, surveying-based acquisition of the fingerprint map, and sufficiently accurate ground truth data Distance error [cm] True distance [m] A set of UWB ranging des and MATLAB libraries for their configuration and operation are available for the experimental investigations within this thesis. If both thesis topics relating to UWB indoor localization are chosen, the two students should cooperate during the data collection for the fingerprint map generation. 2

3 Lorenz Schmid, Dr. David Salido Monzú, Model based improvement of UWB-based indoor localization Ultra-wideband (UWB) ranging may be a highly attractive techlogical basis to develop indoor localization solutions including positioning on construction sites. The time resolution of these radio-frequency signals enables precise delay-based distance measurements to multiple des, which can in turn be used for localization. However, the accuracy of the resulting position estimates is significantly degraded in case of multipath interference due to signal reflections which are typical in complex indoor environments with many rooms and objects. Furthermore, the UWB signals are attenuated by structures like walls and furniture blocking the line-of-sight between the transmitters and receivers of the system. The goals of this thesis are (i) (ii) (iii) to quantify the effect of walls and furniture blocking the line-of-sight both in terms of delay and loss of signal power; to investigate whether it is feasible and useful (for UWB rangingbased indoor localization) to reduce this effect using a floor plan or BIM (e.g. by iteratively estimating corrections based on the number of walls penetrated by the signal); to investigate whether the accuracy of UWB ranging-based indoor localization can be improved by range corrections iteratively interpolated from a map. The investigation shall be based on extensive test measurements carried out within the HIL building, on existing floorplans of the building, on surveyingbased acquisition of range correction maps, and on sufficiently accurate ground truth data. A set of UWB ranging des and MATLAB libraries for their configuration and operation are available for the experimental investigations within this thesis. If both thesis topics relating to UWB indoor localization are chosen, the two students should cooperate during the data collection for the fingerprint map generation. 3

4 Jemil Butt, Robust 2d and 3d Helmert transformation with uncertainties in both coordinate frames The parameters of a Helmert transformation are usually estimated within a Gauss-Markov model i.e., neglecting the uncertainties in one of the two coordinate frames. However, this is only justified if one of the two frames is much more accurate than the other one. This assumption is w often unjustified since coordinate frames are increasingly of similar accuracy. In this master s thesis an approach based on a Gauss-Helmert model (GHM) allowing for full variancecovariance (vc) information in both frames is to be developed and implemented for the 2d and 3d cases using MATLAB. The algorithm shall be robustified such that it can cope with a reasonable fraction of outliers in the given data. The student shall choose and analyze a robustification strategy (e.g., L1 rm minimization, or outlier detection, identification and adaptation (DIA), a RANSAC-like approach, or a combination). The investigation shall include Monte Carlo simulations demonstrating the validity of the chosen approach and implementation, highlighting the impact of correct handling of the uncertainties and the successful robustification. Finally, the practical relevance of the approach shall be assessed taking into account realistic application cases requiring a Helmert transformation. 4

5 Dr. David Salido Monzú, Approaches and tools for geodetic research data management: the example of terrestrial laser scanning The amount of raw and processed data generated in experimental sciences and engineering is increasing tremendously. Managing these data in a way that allows their efficient reutilization whenever need arises is a significant challenge requiring to balance ecomic criteria, technical aspects, and regulations imposed e.g., by customers, laws or business processes. In the academic environment, establishing an appropriate data management scheme is particularly challenging: data may be very heterogeneous, produced by a variety of different sensors, often using custom-made software and one-of-a-kind instruments. They may be acquired and processed by scientists who only work at a particular institution or position for a short time. Any data used to produce scientific output (e.g., publications) needs to stay accessible and useable for a long term such that the research can be reproduced, scrutinized or continued. Often it is very time-consuming to find and newly access previously used data, in particular if the person who collected and stored them in the first place is t there any more. Soon-to-be established open research data policies require redefining data management strategies and ensuring that research output, associated data and metadata are findable, accessible, interoperable and re-usable. Within this thesis the student shall analyze this problem in the geodetic context, focusing on strategies and solutions for managing large amounts of TLS point cloud data. The study should include a survey (i) of data management solutions utilized by institutions and companies with needs comparable to those of GSEG, (ii) of available data management tools, and (iii) of regulations relevant for an ETH unit. Based on this survey, proposals for TLS data management at GSEG including raw data storage, metadata structure and accessibility shall be developed, compared in terms of relevant criteria, and ideally demonstrated prototypically. The approaches shall be reasonable and focus on datasets acquired in the future (retrofitting existing data is a potential advantage but t a requirement). 5

6 Sukant Chaudhry, Modelling and detecting mixed pixels in laser scanning If more than one reflecting surface is illuminated by a laser beam used for distance measurement, the resulting observation may be biased. The existance and size of the bias depends on the distance measurement techlogy (e.g. continuous wave phase measurement, puls-based time-of-flight measurement, full-waveform measurement), properties of the instrument (e.g., laser beam profile, modulation wavelength), surface reflectivities and relative location of instrument, reflectors and footprint. If the instrument cant detect the multi-reflector situation and distinguish the distances, a significant distance error may result. This effect is called mixed pixels effect in laser scanning. It is a common error source contaminating TLS point cloud data and causes apparent, shadow-like replicas of object edges at discrete locations for phasebased scanners, and elongated tail-like artifacts for pulse-based ones. This thesis shall advance the understanding of the effect in terms of driving parameters, minimum and maximum errors, and relation between the discrete locations of the shadow-like replicas and the modulation wavelengths of the scanner. It shall also explore potential approaches to identifying and possibly mitigating mixed pixels automatically within TLS point clouds, analyzing the minimum size and potential further conditions of mixed pixels to be detected. The investigation shall be based on numerical models implemented in Matlab, and on carefully designed experiments preferably carried out in the geodetic metrology lab of the institute. The focus shall be on phase-based measurements. Optionally the investigation can also include puls-based measurements or be refined in order to take the energy distribution within the beam into account, using simulation tools already developed within a PhD thesis (Sukant Chaudhry) and executed on ETH s Euler cluster. 6