SEED Dataset

Experiment Setup

We develop two datasets, SEED-VLA (SEED-Vigilance-Laboratory) and SEED-VRW (SEED-Vigilance-Real-World), for fatigued driving detection based on dry electrode EEG signals acquired in laboratory and real-world scenarios, respectively.

The laboratory experiments are conducted in the controlled environment of the BCMI Laboratory, which is part of the School of Electronic Information and Electrical Engineering at Shanghai Jiao Tong University. The experimental setup primarily comprises an indoor black car and a large electronic screen, which is assembled from nine smaller screens and situated approximately two meters in front of the windshield of the car. The driver's seat of the car is fitted with a steering wheel and foot brake accelerator pedals, enabling driving simulations. As shown in the figure below, virtual driving software can mimic a variety of intricate road and weather conditions, and when combined with the steering wheel and pedals in the black car, it offers the participant a realistic driving sensation. We enforce an upper speed limit for the vehicle to ensure slow-paced driving. During the experiment, we instruct the participants to maintain their focus and silence, facilitating an easy transition into a state of fatigue.

In our real-world driving experiment, we utilize a Benben EV200, a four-seated electric vehicle. To ensure safety, we have the subjects sit in the passenger seat while an experienced driver handles the vehicle. To simulate an authentic driving experience for the subjects, a Logitech gaming steering wheel and foot brake accelerator pedals are installed in the passenger seat, as depicted in the figure below. These devices are powered via the internal cigarette lighter connection of the car to offer a realistic damping sensation.

We activate the double flash and alarm buzzer of the experimental car when driving. We also limit the vehicle speed to a maximum of thirty kilometers per hour to mitigate the risk of accidents. The personal safety of the subjects and drivers is covered by insurance funded by the laboratory. We predetermine the driving route, which runs along the outermost road of the Minhang Campus of Shanghai Jiao Tong University. This route, which is primarily straight and has minimal pedestrian traffic, allows for a safer and more authentic simulation of real driving conditions.

Throughout the driving process, we instruct the subjects to remain relatively still, focusing ahead, with their hands on the steering wheel and their feet on the pedals at all times. They are directed to react in real time to driving conditions, turning the steering wheel in response to upcoming bends or stepping on the brake when pedestrians cross their path.

Subjects

In the laboratory driving experiment, we recruit 20 participants (12 females, average age 21.6 years), while in the real-world driving experiment, we recruit 14 participants (9 females, average age 20.9 years). All the subjects are right-handed and have a habit of napping. None of the subjects consumed alcohol or medication the day before the experiment, and all had sufficient sleep. Before the experiment commences, the subjects are fully briefed on the experimental procedure, and all take a test drive. The experiments are conducted after meals, when drowsiness is usually induced. The duration of participation in the fatigue driving experiment varies for each participant based on their individual circumstances.

Collection Devices

We use a DSI-24 dry electrode EEG cap to collect EEG data when the participants are taking part in the experiments. The sensors of this device are capable of functioning through regular hair without any prerequisite skin preparation. Furthermore, the absence of a need for conductive gel to foster electrical contact with the scalp makes this approach compatible with real-world car driving scenarios. The sensor layout conforms to the International 10-20 system, and the sensors are positioned at Fp1, Fp2, F7, F3, F4, Fz, F8, T3, C3, C4, Cz, T4, T5, the P3, P4, T6, O1, and O2 (18 electrodes, with a sampling rate of 300 Hz). The reference sensor is placed in the standard Pz position.

PERCLOS

We use the eye-tracking glasses produced by the SMI company to capture the subjects' eye movement data. Based on the recorded instances of blinking, scanning, and gazing, we calculate the PERCLOS scores during the driving process as an indicator. The SMI eye-tracking glasses provide information about eye closure, and thus, we can simply calculate the PERCLOS labels by the following formulation:

Dataset Summary

1. EEG: This part includes 20 files of preprocessed raw EEG signals in .edf format, collected from laboratory settings with a sampling rate of 300Hz. To protect privacy, subjects' names have been removed.
2. perclos: It comprises 20 files of continuous vigilance labels, calculated from eye-tracking data and ranging between 0 to 1. To ensure privacy, subjects' names have been anonymized.

1. EEG: Contains 14 files of preprocessed raw EEG signals in .edf format, gathered from real-world environments, with a sampling rate of 300Hz. Participants' names have been omitted to protect their privacy.
2. perclos: Includes 14 files of continuous vigilance labels derived from eye-tracking data, with values between 0 to 1. The names of the participants have been removed for privacy reasons.

Download

Download SEED-VLA and SEED-VRW

Reference

If you feel that the dataset is helpful for your study, please add the following references to your publications.

1. Yun Luo, Wei Liu, Hanqi Li, Yong Lu and Bao-Liang Lu, A cross-scenario and cross-subject domain adaptation method for driving fatigue detection. Journal of Neural Engineering, 21(4): 046004, 2024. [link] [BibTex]