Abstract:

A novel approach was proposed to recognize driver fatigue status under real driving conditions based on approximate-entropy (ApEn) and complexity of steering wheel angle (SWA). This approach first calculated a non-linear ApEn and complexity of SWA in a short time sliding window based on structural theory of
dynamic time-series data analysis. Then a hard threshold based multi-level fatigue status detection model was designed to process the non-linear feature from first stage, and the pre-diagnosis of fatigue level was outputted from a two-input-two-output fatigue mapping rule based on observed samples. The pre-diagnosed fatigue statuses of the observed sample and its adjacent samples were compared to determine the three-level fatigue status of the observed sample. The experiment was set on real road driving environment. The results show that the approach performs an averaged accuracy of 84.6% for three-level fatigue detection on real road driving test. Therefore, the approach can be especially applied to several industrial applications.

Key words: automotive safety, fatigue recognition, steering wheel angle (SWA), approximate entropy (ApEn), complexity