The invention discloses a fatigue driving detection method based on multi-modal information fusion. The method comprises the steps: collecting electroencephalogram signals, facial images and vehicle driving lane images at the same time, constructing and training convolutional neural networks respectively, collecting and inputting to-be-detected data into the trained convolutional neural networks for preliminary prediction, and judging whether a driver is in a fatigue state or not by integrating preliminary prediction results. The method comprises the following specific steps: generating a training set, respectively constructing and training electroencephalogram signal classification, positioning eye and mouth areas, judging opening and closing states of eyes and mouth, positioning a convolutional neural network of lane line positions, collecting to-be-detected data, preliminarily predicting a fatigue driving state, and fusing and judging a fatigue state of a driver. According to the method, the problems of single driver fatigue state judgment basis mode and incomplete representation of the driver fatigue state are solved, the method has good robustness and stability, and the fatigue driving detection precision is improved.