Phase spread segmentation of pedestrians in far infrared images

Abstract

This article describes a methodology for extracting interesting areas in far infrared (FIR) images that may contain pedestrians. It is part of a larger set of algorithms that are part of an Advanced Driver Assistance System (ADAS). The grey level of an object in a FIR image can shift due to changes of the sensor's temperature. In this paper a contrast and luminance invariant method based on the phase congruency of the signal is proposed. The image is exhaustively searched for regions that may contain a pedestrian based on local phase symmetry at different scales and orientation. Areas with high probability are then feed to a subsequent classification step. By applying this method large areas of the image can be safely ignored, reducing the computation time of the classifier. This method has been tested in the IVVI experimental vehicle in real urban driving scenarios. © Springer-Verlag Berlin Heidelberg 2011.