Shape-based Pedestrian Detection in Infrared Images

With the increase of requirement for improving the safety of night driving, automatic pedestrian detection has received more and more attraction. This paper mainly introduces a pedestrian detection method in infrared images. Base on the properties of infrared images, we present a two-step pedestrian detection method including pedestrian candidate selection and validation. The first step is localization of pedestrian candidate, which is to detect warm objects with specific size and aspect ratio. Then, the validation process is based on template matching, which uses multiscale representation and Dynamic Programming for matching deformed and possible occluded contour. The superiority of the proposed shape matching algorithm to the conventional methods is due to the use of hierarchy of segmented representation. It can adjust automatically while the quantity of noise and deformation changes, which improves the accuracy of pedestrian detection. Experimental results have confirmed the effectiveness of the proposed method.