Night-vehicle-mounted infrared pedestrian detection is of great value.It was found that the traditional system is always complicated in structure,and the real-time performance of the pedestrian detection algorithm is difficult to guarantee.To overcome this problem,an uncooled infrared thermal imaging pedestrian detection system based on the FPGA & Davinci-device architecture was presented.High quality preprocessed images can be obtained with the rational use of infrared focal plane.During the ROI isolation,the methods of local adaptive threshold segmentation and morphological process were used to remove the interference of strong background.In addition,the algorithm differentiated the candidate regions with AdaBoost classifier,which was trained with Haar-like features.Experimental results show that the multi-core heterogeneous system has the advantages including compact structure,low data communication difficulty,and strong algorithm portability;FPGA accelerates the algorithm by about 38 %,and it performs in real-time at a rate of 25 f/s.