ZaVI - State Estimation Solely based on Prior Knowledge and Inertial Sensing
Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
Final Report Abstract
Inertial sensors work like the human vestibular system. They can perceive movement in much the same way as we humans with closed eyes. We find our way around for a short time, but the longer we move, the more unsure we become of where we are. We then often use previously collected environment knowledge. For example, if we touch a wall or furniture, we know where we are via the "map in our heads". Therefore, in this project we have investigated how an inertial sensor can locate itself with prior knowledge of the environment. This is advantageous because inertial sensors are cheap and built into every smartphone, while GPS as a particular alternative does not work indoors. Possible applications are e.g. navigation in department stores or a museum guide app that automatically explains the current exhibit. Amateur sports could also benefit from cost-effective motion analysis using smartwatches. To make this possible, we have researched a method for processing prior knowledge. These so-called pseudo measurements act like a sensor. An example: In the inertial sensor’s motion data we recognize that we are taking an elevator. From the building map we know the elevator’s position. Together this is position information and is processed it like a real position measurement. We examined this map-like knowledge in a bouldering application. We recognize gripping a boulder hold in the inertial data. Then the sensor processes the pseudo measurement "I am on a grip". We showed that the sensor can be located with an accuracy of 15cm (median), enough e.g. for sports analysis. The ground on which we move also reveals something about our position. In track cycling, for example, we can locate the bicycle with an inertial sensor and a height map of the track. Basically, the inertial sensor measures the motion over the last few meters. But it doesn't directly measure where you've been or in which direction you're facing. However, the track’s shape limits possible motions because the wheels stay on the track surface. We use this prior knowledge as a pseudo measurement: "The distance from bike to track is 0". If the bike now takes a curve, the sensor knows that it must be on a curved section of the track. In summary, in this project we have converted map-like prior knowledge into a format that can be used by computers and developed methods for combining it with inertial sensor data.
Publications
-
“State Observability through Prior Knowledge A Conceptional Paradigm in Inertial Sensing,” International Conference on Informatics in Control, Automation and Robotics (Position paper), 2019
T. L. Koller, T. Laue, and U. Frese
-
“State Observability through Prior Knowledge: Tracking Track Cyclers with Inertial Sensors,” 2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Sep. 2019
T. L. Koller and U. Frese
-
“State Observability through Prior Knowledge: Analysis of the Height Map Prior for Track Cycling,” Sensors, vol. 20, no. 9, Art. no. 9, Jan. 2020
T. L. Koller and U. Frese
-
“The Interacting Multiple Model Filter on Boxplus-Manifolds,” in 2020 IEEE International Conference on Multisensor Fusion and Integration for Intelligent Systems (MFI), Sep. 2020, pp. 88–93
T. L. Koller and U. Frese
-
“The Interacting Multiple Model Filter and Smoother on Boxplus-Manifolds,” Sensors, vol. 21, no. 12, Art. no. 12, Jan. 2021
T. L. Koller and U. Frese
-
“Event-Domain Knowledge in Inertial Sensor Based State Estimation of Human Motion.” Fusion, 2022, Linköping
T. L. Koller, T. Laue, and U. Frese
