Development of Three Driver State Detection Models from Driving Information Using Vehicle Simulator; Normal, Drowsy and Drunk Driving

Detection of drivers' states is the essential technology not only to prevent car accidents related with their state but to develop self-driving car. Detecting technology generally uses two types of methods; physiological measures and vehicle-based measures. Vehicle-based measures have advantages compared to physiological method such as non-additional device, unsophisticated process and less computational power. For these reasons, vehicle-based measures are used for this study to build the detection system about 3 states; normal, drowsy and drunk driving. In order to achieve this purpose, three types of algorithm models are suggested using vehicle simulator experiments with twelve participants on three states; normal, drowsy and drunk. By analyzing the accuracy of each input packet data combination, the feature values, the configuration of the input data calculated through the vehicle driving data is used to derive the influential factors for predicting the driver state. The results of the models indicate high accuracy and give the possibility to be applied on detecting 3 states in real driving vehicles with the system using combination of developed models.