Location-based services for mobile devices are a recent and growing area of interest, particularly in indoor environments. Several indoor localization solutions relying on smartphones and inertial sensing-based motion tracking algorithms are currently emerging. However, despite being attractive, these types of solutions still present some challenges, including, heading drift, step length estimation errors, missed steps, inaccurate evaluation of unpredictable motion patterns, and consequently, cumulative errors and long-term inaccuracy.
Often, forward filtering-based methods are applied to mitigate some of these problems. However, unless environmental reference technologies (e.g. magnetic and WiFi signals) are used, errors will be accumulating over time, without the possibility to recover/re-calibrate.
However, due to the intrinsic characteristics of human walking and because of metric, topological and semantic constraints imposed by indoor maps, human walking movements are highly structured and can be described using a relatively small vocabulary of motions.
Using this knowledge, more effective map matching algorithms can be applied, which can then be used to compensate errors, auto adapt and learn motion parameters and recover from unpredictable motion patterns.
The aim of this project is therefore to develop a set of adaptive strategies for motion self-learning, based on forward-backward correction mechanisms, capable of improving steps detection, step length and heading estimation robustness. Moreover, through the application of trajectory matching algorithms, and based on the constraints imposed by indoor maps and motion compatibility, the developed system must be able to automatically re-calibrate in case the position is lost and recover the user’s recent path history.
For any additional information regarding this project, please contact us using the inquiries form.