Challenges of Large-Scale Multi-Camera Datasets for Driver Monitoring Systems.

Area of Science:

• Computer Science
• Artificial Intelligence
• Robotics

Background:

• Deep neural networks (DNN) and Big Data (BD) have driven significant progress in advanced driver assistance systems (ADAS).
• Large, complex datasets are crucial for benchmarking and advancing research and development in the ADAS/AD domain.
• Multi-modal datasets enhance DNN models by fusing information from diverse sensors, optimizing performance through redundancy and complementarity.

Purpose of the Study:

• To explore the requirements and technical approach for building a multi-sensor, multi-modal dataset for video-based ADAS/AD applications.
• To introduce the Driver Monitoring Dataset (DMD) and its partial release to promote research in driver monitoring systems (DMS).

Main Methods:

• Exploration of Big Data dimensions: volume, variety, veracity, visualization, and value.
• Development of a multi-sensor, multi-modal dataset tailored for video-based ADAS/AD.
• Detailed presentation of dataset preparation, construction, post-processing, labeling, and publication strategies.

Main Results:

• The creation and partial release of the Driver Monitoring Dataset (DMD).
• The DMD is designed to address the specific needs of driver monitoring systems (DMS) research.
• A subsequent public release of DMD materials is announced to further community engagement.

Conclusions:

• The development of specialized, multi-modal datasets is essential for advancing ADAS/AD technologies.
• The Driver Monitoring Dataset (DMD) provides a valuable resource for the research community, particularly for driver monitoring systems.
• Open access to such datasets accelerates innovation and development in autonomous driving systems.