| Summary: | Rider’s gaze data may contain a rich source of information on a rider’s intent to overtake. This relationship has been widely studied for car drivers. To enhance the safety of Personal Mobility Device (PMD) riders and pedestrians in their paths, this project targets to predict the direction of turning manoeuvres of a PMD rider when a pedestrian is in front of the rider, by analysing the PMD rider’s gaze data. Gaze features including gaze accumulation and glance frequency are first formulated, then a machine learning model is trained to find the relationship between the gaze features and the rider’s turning manoeuvres. Besides predicting the direction of turning manoeuvres, whether a pedestrian is in front of the rider is also determined with the aim of completeness of the algorithm. Compared with similar studies on car drivers, the proposed framework for PMD riders adopts fewer data types and contains compensation techniques for noisy/missing data points frequently encountered in outdoor measurement. Real-time experiments are carried out and the results show that through our proposed technique can predict the PMD rider’s direction of turning manoeuvres with an accuracy of 96%, 1 second in advance. This shows the feasibility of turning manoeuvres prediction based on gaze data and serves as an inspiration for developing a complete ADAS (Advanced driver-assistance systems) for PMD riders.
|
|---|