Fraunhofer Center for Assistive Information and Communication Solutions – AICOSDescription:
People across the world are no strangers to poor information in unknown subway networks, which cause delays or mistakes in their trajectory. A thorough research reveals that underground navigation solutions are scarce and mainly for academic purposes. Insufficient demand coupled with the limitations of existing navigational technology in the environment, such as GNSS, are the main reasons for this scarcity on a global level. However, this poses as an opportunity to provide additional information to tourists and natives, improving the user experience in subway networks.
To achieve this solution, SubwayNav tackles navigation in these environments using digital sensors which are present on most Android mobile devices, such as the accelerometer, agnetometer and gyroscope. Collaborator in the field of urban mobility, Ubirider, proposed the main functionalities which include a stop counter, direction detector and a simple route planner, relying solely on smartphone sensors and emphasizing adaptability, precision and efficiency.
Counting stops requires distinguishing between at least two states: moving within a metro, and stopped in any other situation. Preliminarily, considering this a classification problem, sensor data is repeatedly acquired in metro trips, extensively analyzed and processed to identify the features that best distinguish the states from each other. Resorting to supervised machine learning algorithms, trained with the best performing features for better robustness, the mobile device may feed short windows of sensor data in real time, thus keeping track of its state. To detect direction, time between stops is compared to a time window ranging between fixed margins before and after the average duration to the correct station. With this information, users are likely to reach their destination safely and on time. Finally, a visually basic proof of concept was developed and tested in the field to assess the performance, value and reliability of the developed algorithm.
Author: Afonso Vogensen
Type: MSc thesis
Partners: Instituto Superior de Engenharia do Porto; Ubirider SA
Year: 2019