Night is the period of fatigue for drivers and of accidents happened. At night, Ambient lighting is weak, and the full-in light of visible light influence the normal driving. According to these conditions, a infrared active vision fatigue detection design for night driving is proposed . Firstly,the 940nm light source with no red is adopted to perform fill-in light,by which imaging quality is developed and the interference of ambient light is prevented. Secondly, the existing fatigue detection algorithm is improved. Face location and tracing is realized with ST_Adaboost and KLT tracing. Eyes are located by gray-level integration method for Gabor transformation,which increases the accuracy of eye positioning. Finally, the fatigue is judged by PERCLOS algorithm. The experiment results show that the proposed algorithm can detect fatigue driving around the clock, and the accuracy of eye positioning is more than 90%, the accuracy of eye state recognition is more than 87%. The proposed method can resolve the problem about driving fatigue at night, so there is great significance for lowering the risk of traffic accidents especially at night.