A visual perception approach for accurate segmentation of light profiles

Abstract

In this paper we present the automatic real time segmentation algorithm we devised to be consistent with human visual perception for a highly contrasted scene, like the one generated by the projection of the luminous profiles from high power sources on a uniform untextured pattern. An accurate identification of shadow-light profiles is required, for example, from industrial diagnostics of light sources, in compliance with regulations for their employment by human users. Off-the-shelf CCD technology, though it could not be able to cover the wide dynamic range of such scenes, could be successfully employed for the geometric characterization of these profiles. A locally adaptive segmentation algorithm based on low-level visual perception mechanisms has been devised and tested in a very representative case study, i.e the geometrical characterization of beam profiles of high power headlamps. The evaluation of our method has been carried out by comparing (according to a curve metric) the extracted profiles with the ones pointed out by five human operators. The experiments prove that our approach is capable of adapting to a wide range of luminous power, mimicking visual perception correctly even in presence of low SNR for the acquired images. © 2009 Springer Berlin Heidelberg.