Steering wheel motion analysis for detection of the driver's drowsiness

Reliable system for driver's drowsiness recognition is the aim of many studies. Unfortunately, majority of researchers work with data acquired in laboratory with ideal or simulated conditions. Therefore it is difficult to implement their results to real car and prove its reliability and accuracy. The analyzed data in this paper is acquired from real traffic and therefore it contains all disadvantages partially modeled in laboratory. For data acquisition has been chosen in-direct measurement from car CAN bus in order to not affect the driver. All data are preprocessed according to assumptions about driver's behavior and transformed to frequency domain by means of orthogonal transform (STFT, CWT and DWT). Subsequently, data is analyzed by data mining methods including features extraction and filter feature selection. The performance of the features is measured by the area under the receiver operating characteristic.