Abstract: In order to reduce the image detection effect from the changes of the driver’s position in the driver fatigue detection, a driver fatigue detection method is proposed based on multi-scale local binary pattern histogram Fourier feature (MLBPH-FF) and support vector machine (SVM). The method includes two processes which are training and recognition. During training, we firstly extract the features of the driver’s facial fatigue and non-fatigue images captured from video stream, calculate and get the MLBP histogram (MLBPH) using different scales of the uniform local binary pattern (LBP) operators, then combine them and use discrete Fourier transform to get the MLBPH-FF. At last, we input these features data to the SVM and train them to get its model and parameters. During recognition, we calculate the MLBPH-FF of the testing image samples, then input these MLBPH-FF to the trained SVM to detect fatigue. The experimental result shows that this method is of better identification rate on fatigue detection and performs stably and robustly on different illuminations and poses.

Key words: fatigue detection, MLBPH-FF, SVM