Skin segmentation using a multiband camera for early pedestrian detection

Advanced driver assistance systems require fast and robust algorithms for understanding complete traffic scenarios that include vehicles, pedestrians, and road boundaries. Although conventional approaches based on pattern recognition is powerful and effective for achieving high performance, they have a large amount of computational burden. Therefore, we have been developing an object recognition technique based on spectroscopy. All substances have their own reflection characteristics, and human skin in particular has a unique distinction in the visible and near-infrared (NIR) regions. Therefore, a multiband camera, which can simultaneously obtain seven spectral images, was developed to realize early pedestrian detection. In this paper, two approaches are presented to detect human skin from spectral images in an outdoor environment. One is an analytical method based on visible color information and subtraction between the NIR spectral images. Another is a statistical method for learning the brightness distribution of human skin in the seven spectral images. The experimental results confirm the ability of the multiband camera to detect skin. In addition, some problems and improvements of the proposed methods in an outdoor environment are discussed.