Accurate Estimation of Indoor Travel Times - Learned Unsupervised from Position Traces

The ability to accurately estimate indoor travel times is crucial for enabling improvements within application areas such as indoor navigation, logistics for mobile workers, and facility management. In this paper, we study the challenges inherent in indoor travel time estimation, and we propose the InTraTime method for accurately estimating indoor travel times via mining of historical and real-time indoor position traces. The method learns during operation both travel routes, travel times and their respective likelihood - both for routes traveled as well as for sub-routes thereof. InTraTime allows to specify temporal and other query parameters, such as time-of-day, day-of-week or the identity of the traveling individual. As input the method is designed to take generic position traces and is thus interoperable with a variety of indoor positioning systems. The method's advantages include a minimal-effort setup and self-improving operations due to unsupervised learning - as it is able to adapt implicitly to factors influencing indoor travel times such as elevators, rotating doors or changes in building layout. We evaluate and compare the proposed InTraTime method to indoor adaptions of travel time estimation methods for outdoor navigation. Our extensive evaluation uses datasets collected in real-world hospital work environments. InTraTime is deployed at a hospital as an online system, demonstrating that it learns automatically and in real-time travel times as position traces are collected within the building complex. Results indicate that InTraTime is superior with respect to metrics such as deployment cost, maintenance cost and estimation accuracy, yielding an average deviation from actual travel times of 11.7%. This accuracy was achieved despite using a minimal-effort setup and a low-accuracy positioning system.