TD
T.T. Dollé
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
1 records found
1
Seeing Future Sickness
Vision-Based Head Pose Estimation for Motion Sickness Assessment in Automated Vehicles
Automated Vehicles (AVs) may improve road safety, traffic efficiency, and occupant productivity by enabling passengers to engage in Non-Driving Related Tasks (NDRTs). However, reduced control, limited anticipation of vehicle motion, and visual distraction can exacerbate Motion Sickness (MS). This work proposes a non-intrusive computer vision framework for estimating passenger head motion from depth-camera data to support real-time MS assessment.
A machine-learning-ready RGB-D dataset was created using a custom 6-DoF motion simulator, with an Azure Kinect depth camera and motion capture system providing synchronised measurements and ground-truth poses. An anatomically grounded head coordinate system based on the vestibular organs was introduced to align the measurements with human motion perception. The dataset comprises 42 trials from 17 participants and 323,424 frames, covering different driving speeds, glasses conditions, and instructed head movements.
Five machine learning architectures were evaluated using 5-fold participant-level cross-validation. A post-processing pipeline combining coordinate calibration and non-causal low-pass filtering reduced tracking errors by approximately 2.5–4 times across all models. The best-performing model achieved mean MSDV-weighted coherence values of 0.890 for orientation and 0.941 for position within the primary MS frequency range (0.05–1 Hz). These results demonstrate the feasibility of depth-camera-based head-motion measurement for mathematical MS prediction and non-intrusive passenger comfort monitoring in future AVs. ...
A machine-learning-ready RGB-D dataset was created using a custom 6-DoF motion simulator, with an Azure Kinect depth camera and motion capture system providing synchronised measurements and ground-truth poses. An anatomically grounded head coordinate system based on the vestibular organs was introduced to align the measurements with human motion perception. The dataset comprises 42 trials from 17 participants and 323,424 frames, covering different driving speeds, glasses conditions, and instructed head movements.
Five machine learning architectures were evaluated using 5-fold participant-level cross-validation. A post-processing pipeline combining coordinate calibration and non-causal low-pass filtering reduced tracking errors by approximately 2.5–4 times across all models. The best-performing model achieved mean MSDV-weighted coherence values of 0.890 for orientation and 0.941 for position within the primary MS frequency range (0.05–1 Hz). These results demonstrate the feasibility of depth-camera-based head-motion measurement for mathematical MS prediction and non-intrusive passenger comfort monitoring in future AVs. ...
Automated Vehicles (AVs) may improve road safety, traffic efficiency, and occupant productivity by enabling passengers to engage in Non-Driving Related Tasks (NDRTs). However, reduced control, limited anticipation of vehicle motion, and visual distraction can exacerbate Motion Sickness (MS). This work proposes a non-intrusive computer vision framework for estimating passenger head motion from depth-camera data to support real-time MS assessment.
A machine-learning-ready RGB-D dataset was created using a custom 6-DoF motion simulator, with an Azure Kinect depth camera and motion capture system providing synchronised measurements and ground-truth poses. An anatomically grounded head coordinate system based on the vestibular organs was introduced to align the measurements with human motion perception. The dataset comprises 42 trials from 17 participants and 323,424 frames, covering different driving speeds, glasses conditions, and instructed head movements.
Five machine learning architectures were evaluated using 5-fold participant-level cross-validation. A post-processing pipeline combining coordinate calibration and non-causal low-pass filtering reduced tracking errors by approximately 2.5–4 times across all models. The best-performing model achieved mean MSDV-weighted coherence values of 0.890 for orientation and 0.941 for position within the primary MS frequency range (0.05–1 Hz). These results demonstrate the feasibility of depth-camera-based head-motion measurement for mathematical MS prediction and non-intrusive passenger comfort monitoring in future AVs.
A machine-learning-ready RGB-D dataset was created using a custom 6-DoF motion simulator, with an Azure Kinect depth camera and motion capture system providing synchronised measurements and ground-truth poses. An anatomically grounded head coordinate system based on the vestibular organs was introduced to align the measurements with human motion perception. The dataset comprises 42 trials from 17 participants and 323,424 frames, covering different driving speeds, glasses conditions, and instructed head movements.
Five machine learning architectures were evaluated using 5-fold participant-level cross-validation. A post-processing pipeline combining coordinate calibration and non-causal low-pass filtering reduced tracking errors by approximately 2.5–4 times across all models. The best-performing model achieved mean MSDV-weighted coherence values of 0.890 for orientation and 0.941 for position within the primary MS frequency range (0.05–1 Hz). These results demonstrate the feasibility of depth-camera-based head-motion measurement for mathematical MS prediction and non-intrusive passenger comfort monitoring in future AVs.