Vision-based protective devices. Parametric analysis of image descriptors detecting changes in a known texture (pattern)

The progress made in recent years in the area of digital vision systems now leaves more scope for designing human detection devices based on this technology.
One of the main difficulties for the design of such a device is linked to the need to identify one or several characteristics (dimensions, shape, texture, colour, movement, etc.) in an image (or a sequence of images) which translate the presence or absence of a person in an industrial environment likely to vary strongly.
Based on the hypothesis that the zone to be protected is covered by contrasted periodic pattern, this paper presents a set of methods to extract contour, colour and texture information, and assesses their limits of use in relation to the safety of machinery.
In particular, it has been demonstrated that detection algorithms based on contours (strong transitions in the image) as well as those based on colour information significantly contribute to improving the robustness of the detection in relation to luminous interference. These algorithms could therefore ensure the detection function (availability) in conditions of acceptable lighting.