Visual attention is an important factor when studying driver behavior. While the location of the pupil can provide accurate information about gaze, the vehicle environment poses challenges that prevent the use of off-the-shelf gaze detection algorithms in the car. Head pose can be used to approximate the driver's visual attention, providing a coarse estimate which can be good enough for many applications. However, the relation between head pose and gaze is not one-to-one, depending on the driver, cognitive load, and visual task. Instead of detecting a precise gaze direction, this paper proposes a novel approach which creates a probabilistic map describing visual attention. The approach relies on Gaussian process regression (GPR), which takes the position and orientation of the drivers' head to estimate the probability that the driver is looking at a given point. The approach creates confidence regions describing the most likely gaze directions. We evaluate the proposed approach with naturalistic recordings in real roads, where we estimate the position and orientation of the driver's head using a headband with fiducial markers. The experimental evaluation demonstrates that 89.4% of drivers' gaze are included in the 95% confidence region predicted by our model. The proposed system can provide valuable information for navigation, infotainment, safety and communication systems.
Probabilistic estimation of the driver's gaze from head orientation and position
01.10.2017
1352214 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
IEEE | 2023
|Real-Time Gaze Estimator Based on Driver's Head Orientation for Forward Collision Warning System
Online Contents | 2011
|Europäisches Patentamt | 2024
|Europäisches Patentamt | 2024