Multidomain Fusion-Based Fatigued Driving Detection Using Frequency-Modulated Continuous Wave Radar

Facial movements are one of the key indicators of a driver’s state. Accurate estimation of facial movements can improve the accuracy of fatigue driving detection. The existing solutions suffer from inconvenient configuration or privacy leakage risks. Radar offers inherent privacy protection and non-contact operation and holding the potential to address the aforementioned shortcomings. Therefore, this paper proposes a method for facial movement recognition using a Frequency-Modulated Continuous Wave (FMCW) radar. First, an exponential weighting cancellation algorithm is applied to suppress interference and enhance the signal-to-noise ratio of the facial signal, thereby enabling further extraction of the target’s amplitude signal. Then, the intermediate frequency (IF) signal is decomposed using Multivariate Variational Mode Decomposition (MVMD), and a correlation-based selection criterion is proposed to select potential eigenmodes for blinking and yawning signals construction. The micro-Doppler features are obtained via Short-Time Fourier Transform (STFT). Finally, we propose the Multi-scale information fusion Convolutional neural network for Small sample-Driven Fatigue detection (MCSF-Net), which integrates the target’s frequency-domain micro-Doppler features with time-domain amplitude signals for fatigue detection. Experimental results show that the average classification accuracy reaches 92.71 ± 2.13, demonstrating the effectiveness of the proposed method.