Segmentation of pedestrians in thermal imaging

Abstract

Image segmentation is one of crucial stages of video processing used for detection of pedestrians in infrared (IR) vision. The paper presents a study of passive (typically far IR) technologies for IR vision and an analysis of the state-of-the-art in the extraction of information from IR images. A deep testing of segmentation methods was performed for IR images with a specially prepared benchmark of 162 IR video frames with pedestrians. The initial tests included two segmentation methods: the single threshold and the Otsu method. The obtained results are promising as the effectiveness reaches ca. 82% for both techniques.