In order to solve the problem of slow recognition speed ow accuracy of traditional pedestrian detection methods in near-infrared（IR） night pedestrian detection

an improved Faster R-CNN night-IR pedestrian detection method is proposed.In order to extract deeper image information

feature extraction network is designed as ResNet-50.Aspect ratio of target is calculated

and the weighted anchor frame mechanism with target scale is introduced.Feature map fusion is performed on multi-level feature map.The experimental results on the self-built night near-IR pedestrian dataset show that compared with the traditional LBP+Haar+HOG+SVM algorithm and the classic Faster R-CNN algorithm

mean average precision（MAP） of night near-IR pedestrian detection is increased by 50 % and 11.2 %

and the detection speed is increased from 2000 ms/f and 214 ms/f to 190 ms/f.