Fatigue driving is one of the major causes of traffic accidents. Additionally, the prediction of fatigue state is of crucial significance for pilots, drivers and chief conductors. Nevertheless, most of the existing Fatigue Driving Detection Systems (FDDS) are based on single signal source or extract single feature, which suffer from dim lighting, varying skin color, inaccuracy and high cost. Hence, this thesis is the first to propose a cheap, easy-to-use, accurate FDDS based on video signals, which combines geometry and texture features and predicts fatigue state by combining video with Electroencephalography(EOG) signals and gripping power. The author is the first to propose fatigue driving experiments by combining video and EOG signals on ICONIP 2012: EOG is more accurate while video includes more information; by combing both signals, we greatly improve the prediction accuracy. This system won the 2nd prize on ICCF 2011. Meanwhile, the author conducted similar experiments with gripping power, which again proves the robustness and accuracy.