Negar Alinaghi, Markus Kattenbeck, Antonia Golab, and Ioannis Giannopoulos
Results of machine learning experiments for turning activity classification.
Decision making is an integral part of wayfinding and people progressively use navigation systems to facilitate this task. The primary decision, which is also the main source of navigation error, is about the turning activity, i.e., to decide either to turn left or right or continue straight forward. The fundamental step to deal with this error, before applying any preventive approaches, e.g., providing more information, or any compensatory solutions, e.g., pre-calculating alternative routes, could be to predict and recognize the potential turning activity. This paper aims to address this step by predicting the turning decision of pedestrian wayfinders, before the actual action takes place, using primarily gaze-based features. Applying machine learning methods, the results of the presented experiment demonstrate an overall accuracy of 91% within three seconds before arriving at a decision point. Beyond the application perspective, our findings also shed light on the cognitive processes of decision making as reflected by the wayfinder’s gaze behaviour: incorporating environmental and user-related factors to the model, results in a noticeable change with respect to the importance of visual search features in turn activity recognition.
Leibniz International Proceedings in Informatics (LIPIcs), 11th International Conference on Geographic Information Science (GIScience) - Part 2, 5, 1-16, 2021-09-14.