Learning Multimodal Representations for Drowsiness Detection

Drowsiness detection is a crucial step for safe driving. A plethora of efforts has been invested on using pervasive sensor data (e.g., video, physiology) empowered by machine learning to build an automatic drowsiness system. Nevertheless, most the existing methods are based complicated wearables electroencephalogram) or computer vision algorithms eye state analysis), which makes relevant systems hardly applicable in wild. Furthermore, these insufficient nature due limited simulation experiments. In this light, we propose novel and easily implemented method full non-invasive multimodal analysis driver task. The level was estimated self-reported questionnaire pre-designed protocols. First, consider involving environmental temperature, humidity, illuminance, further more), can be regarded as complementary information human activity recorded via accelerometers actigraphs. Second, demonstrate that models trained daily life still efficient make predictions subject performing simulator, may benefit future collection methods. Finally, comprehensive study investigating different including classic ‘shallow’ recent deep models. Experimental results show that, our proposed reach 64.6% unweighted average recall subject-independent scenario.