Detecting Driver-induced Steering Oscillations through Adaptive Thresholding of the Power Spectrum of Vehicle's Lateral Acceleration

抄録

This study propose an approach for detecting the steering oscillations caused by delayed steering response in a human driver in normal driving (routine) scenarios. The delayed steering response is a result of inadequate cognitive load on distracted or cognitively overloaded drivers. The proposed approach is based on real-time thresholding of the power spectrum magnitude of a Fourier-transformed signal, which is obtained from the lateral acceleration of a car. However, even if the power spectrum magnitude of inattentive driving is higher than that of fully attentive driving in the same driving scenario because of steering oscillations, it varies significantly when the driving scenario changes. Inattentive driving along straightaways yields anomalously lower power spectrum values than those obtained from attentive driving when cornering. Therefore, it would be unfeasible to employ a single threshold to detect steering oscillations across all driving situations. Hence, we propose using the absolute value of the lateral jerking motion of a car as an indication that the car is entering or exiting a corner. The experimental results indicated that adaptive thresholding of the power spectrum of lateral acceleration facilitates the detection of driver-induced steering oscillation with an overall accuracy of 88%.