Abstract: The contrast between pedestrians and environment is low in infrared images,which will result in the problem of faulty segmentation.This paper puts forward a pedestrian segmentation algorithm in infrared images based on frame difference and Otsu.Firstly,two frames with a short time interval of an infrared video sequence are used to obtain a frame difference image,and Otsu method is applied to get some moving target regions in the frame difference image.Then,the reference frame is also segmented by Otsu method,and the seed pixels of target regions can be gained by multiplying the result of the frame difference image segmentation with the result of the reference frame segmentation.Finally,according to region connectivity of the segmentation image of the reference frame,pedestrian areas are determined.Experimental results show that the proposed algorithm can overcome the interference of the environment,has strong robustness,and can effectively realize pedestrian segmentation in the infrared image.

Key words: reference frame, frame difference, Otsu method, alpha-trimmed mean filter, connectivity