Color histogram footprint technique for visual object tracking

Abstract

The use of video cameras for robot control calls for fast algorithms to identify and localize target objects in the video image. We evaluate the possibility of working with daily life objects from the home environment, without any specific markers, using color information. The color histogram is an interesting mean to compare a reference image of an object with its image projection in the scene. We evaluate the adaptation of this approach to visual servoing that must handle the important variations in size and appearance of the object. In this paper, we introduce a new technique: the color histogram footprint, which is a kind of color histogram based threshold. This technique is very fast and is designed to complement a prior coarse object localization phase. It creates a binary image where the outline of the object stands out, like a footprint. It allows thus to determine the precise size and location of the object in the image