Malaysian Driver Behavior Dataset: 7 Action Classes from 44 Participants

The Malaysian Driver Behavior Dataset (MDBD) was developed at the Vehicle Intelligence and Telematics Lab, Universiti Teknologi MARA. It contains over 155,000 high-resolution front-facing images from 44 participants, captured in a driving simulator under day and night conditions to represent 7 driver behavior classes. The dataset was last updated on July 16, 2026.

Use Cases

• Train classifiers to detect yawning and drowsiness based on facial images.
• Develop models robust to lighting variations using the day and night subsets.
• Study naturalistic distracted behaviors like texting and calling without strict posture constraints.
• Analyze the impact of occlusions such as glasses and hijabs on behavior recognition.

Strengths

• Data from 44 participants (26 male, 18 female) with varied ages, genders, and ethnic backgrounds to reduce bias.
• Includes over 155,000 images split into three distinct subsets: Day (53,433), IR (50,216), and Night (52,260).
• Captures 7 specific driver behavior classes, including normal, yawning, drowsy, calling, distracted, texting, and interacting.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Data was collected in a simulated environment, which may not fully capture all real-world driving complexities.

Provenance

Source

Vehicle Intelligence and Telematics Lab, Universiti Teknologi MARA, Selangor, Malaysia.

Collection Method

Collected using a driving simulator from 44 participants across two sessions (day and night).

Freshness

Last updated 2026-07-16 17:37:45; freshness should be verified.

Geography

Malaysia (simulated environment).