The Multimodal Driver Monitoring Database: A Naturalistic Corpus to Study Driver Attention

A smart vehicle should be able to monitor the actions and behaviors of human driver provide critical warnings or intervene when necessary. Recent advancements in deep learning computer vision have shown great promise monitoring activities. While these algorithms work well a controlled environment, naturalistic driving conditions add new challenges such as illumination variations, occlusions extreme head poses. vast amount in-domain data is required train models that high performance predicting related tasks effectively behaviors. Toward building infrastructure, this paper presents multimodal (MDM) dataset, which was collected with 59 subjects were recorded performing various tasks. We use Fi- Cap device continuously tracks movement using fiducial markers, providing frame-based annotations pose conditions. ask look at predetermined gaze locations obtain accurate correlation between driver's facial image visual attention. also collect performs common secondary activities navigation phone operating in-car infotainment system. All are definition RGB cameras time-of-flight depth camera. record controller area network-bus (CAN-Bus), extracting important information. These quality recordings serve ideal resource efficient for driver, further field in-vehicle safety systems.