Robust object finding vision system based on saliency map analysis

This paper describes a new method of constructing a saliency map that can be used in object finding for a robot vision system. A saliency map shows the weight of each pixel in an image. The weights represent the probability of the location of a given object. Traditionally, a saliency map is constructed based on the analysis of the image color features of the RGB color space. We redefine the saliency map of an image on the HSL color space, which makes our method more robust under a wide range of luminance. Experimental results show that the range can cover most cases in everyday life. Thus, the system can successfully find objects under different conditions and various light sources.