The traditional infrared image pedestrian detection method uses artificial scale template to design and extract pedestrian contour feature.Because the preset template proportion is relatively fixed,when the pedestrian's contour proportion changes greatly due to the increase or decrease of clothing,the change of carrying on items and postures,the direction will lead to the algorithm failure and the phenomenon of missing detection.The target detection based on deep learning can realize the learning of more diverse features by abstracting,extracting,processing and integrating the essential features of a large number of samples.Therefore,the research on the application of deep learning target detection algorithm in infrared image pedestrian detection can make up for the shortcomings of traditional detection methods.YOLOv3 is a relatively balanced recognition algorithm at present.Based on the analysis of the principle and characteristics of YOLOv3 series algorithm,this paper proposes a new improved algorithm model,which can improve the detection speed with almost no loss of detection accuracy.Thus,to a certain extent,the relative balance of detection accuracy and speed is achieved.