Infrared stereo vision-based pedestrian detection

Abstract

Abstruct-This paper describes an approach for pedestrian detection in stem infrared images. The developed system has been implemented on an experimental vehicle equipped with two infrared camera and preliiariIy tested in diBerent situations. It is based on the localization and distance estimation of warm areas in the scene; the algorithm groups areas with similar position and considers only results with specific size and aspect ratio. A final validation process, based on the head shape's morphological and thermal characteristics, is used to build the list of potential pedestrian appearing in the scene. Neither temporal correlation, nor motion cues are used in this processing.

Structure from Infrared Stereo Images

“…Some examples include military applications such as target acquisition [20], autonomous vehicle navigation [66], collision avoidance [88], terrain analysis [69], etc. ; and surveillance applications like pedestrian detection/tracking [4], [52], face detection/recognition [43], [77], etc.…”

“…This means that images of objects at different distances from the eyes are projected in the two eyes with different horizontal disparities, giving some cue for depth computation. The same technique has been used in most computational vision applications for three-dimensional structure recovery (e.g., in [4], [21], [27], [45]), usually referred to as stereo computer vision.…”

“…Besides, almost most of the existing techniques for IR stereo matching (like [4], [54], [87]) are based on some assumptions about the hot contents of the images (e.g., humans, faces, etc.) and therefore are not applicable to other situations.…”

Pedestrian Detection in Far Infrared Images based on the use of Probabilistic Templates

“…Far infrared (FIR) cameras have demonstrated to be adequate when the background is colder than the human shape [1], in low illumination conditions or during the night [26]. FIR images generally are less sensitive to noise, easing the detection process, but these approaches can fail if the background is warmer than the pedestrians or if in the scene there are other objects that emit heat.…”

“…where th(x, y) is the thresholded image, I(x, y) is the raw input image and the value of threshold is obtained as explained in (1). The pixels that belong to the background are given a negative value and the pixels that correspond to the foreground are replaced by 1.…”