Description
The "ULg Multimodality Drowsiness Database", also called DROZY, is a database containing various types of drowsiness-related data (signals, images, etc.) and intended to help researchers to carry out experiments, and to develop and evaluate systems (i.e. algorithms), in the area of drowsiness monitoring.
The (multimodality) data were collected by the Laboratory for Signal and Image Exploitation (INTELSIG), which is part of the Department of Electrical Engineering and Computer Science of the University of Liège (ULg), Liège, Belgium. This laboratory is denoted here by "ULg-INTELSIG".
Update (July 2018)
Due to logistical reasons, the DROZY database is now hosted on the Open Repository and Bibliography (ORBi) of the ULg, instead of on a dedicated server. We excluded from DROZY the raw near-infrared intensity and range images because of limited storage space. However, the encoded videos of the intensity images are still included. Furthermore, given that some frames are missing, we added helpful data for interpolating (features of) them.
Content
The data in the DROZY database were collected from 14 young, healthy subjects (3 males, 11 females) who performed three successive 10-min psychomotor vigilance tests (PVTs) in conditions of increasing sleep deprivation induced by acute, prolonged waking. For each subject, and for each PVT, the database contains the following perfectly time-synchronized data:
- Karolinska Sleepiness Scale (KSS) scores: subjective, self-evaluated levels of drowsiness according to the KSS, thus from 1 (Extremely alert) to 9 (Very sleepy, great effort to keep awake, fighting sleep), at the beginning of each PVT;
- PVT data: stimulus times and corresponding reaction times (in milliseconds);
- Polysomnography (PSG) signals: 5 EEG channels (Fz, Cz, Pz, C3, and C4, all referenced to A1), 2 EOG channels (horizontal and vertical), ECG, and EMG, all sampled at 512 Hz; in .edf format;
Kinect v2 sensor images:near-infrared (NIR) intensity and range images with size of 512x424 pixels, grayscale and distance values of 16 bits, and framerate of 30 fps; in .png format;- Kinect v2 sensor videos: near-infrared (NIR) intensity videos (1 per PVT) with size of 512x424 pixels and grayscale values of 8 bits; in .mp4 format;
- Face landmarks (obtained manually): 2D (in pixels) and 3D (in millimeters) annotations of 68 face landmarks for 720 hand-selected frames (averaging 51 per subject), resulting from manual processing;
- Face landmarks (obtained automatically): 2D and 3D annotations of 68 face landmarks for all frames, resulting from automatic processing using subject-specific constrained local models (CLMs);
- (NEW) Interpolation indices: interpolation coefficients and indices of nearest frames, necessary for interpolating informations of missing frames.
Publishable images
Only the images of the following subjects are publishable: 1, 2, 3, 6, 7, 8, 10, 11, 12, 13, 14. While these images can be shown, the identity of the subjects (which we do not provide) cannot be revealed in any way.
Citation
All documents (such as publications, presentations, posters, etc.) that report results, analysis, research, or equivalent that were obtained by using the DROZY database must cite, at a minimum, the following research paper:
"The ULg Multimodality Drowsiness Database (called DROZY) and Examples of Use", by Quentin Massoz, Thomas Langohr, Clémentine François, Jacques G. Verly, Proceedings of the 2016 IEEE Winter Conference on Applications of Computer Vision (WACV 2016), Lake Placid, NY, March 7-10, 2016. [IEEE Xplore] [ORBi] [pdf]
This paper describes the DROZY database, and presents two examples of use, i.e. the development and evaluation of two systems predicting the reaction times based on ocular parameters.
Limitation
The DROZY database is only available to researchers (in universities, research laboratories, or equivalent) for their own research. Specifically, the database is not available to commercial companies for the development of commercial products.
Access
The DROZY database is now available for download on the Open Repository and Bibliography (ORBi) of the University of Liège.