Modeling of photometric variations for matching correspondences, its validity and its application to tracking

Michèle Gouiffès (1) , Christophe Collewet (2)

1 IEF - Institut d'électronique fondamentale
2 Lagadic - Visual servoing in robotics, computer vision, and augmented reality

Since modeling reflections in image processing is a difficult task, most computer vision algorithms assume that objects are Lambertian and that no lighting change occurs. Some photometric models can partly answer this issue by computing the illumination changes in small areas of the image, but they often assume that the lighting changes are the same in each point of a window of interest. Through a study based on specular reflection models, such as the Phong and the Torrance-Sparrow ones, we explain explicitly the assumptions on which these models are implicitly based and therefore the situations in which they fail. In this report, we propose two photometric models, which compensate for specular highlights and lighting variations. They are based on the assumption that illumination changes vary smoothly on the window of interest. The first one is more suitable when specular highlights occur and when small windows of interest are used, as in feature points tracking. The second model compensates for more comprehensive changes such as specular highlights and lighting changes, and can be used on larger areas of the image. Contrary to existing models, the characteristics of the surface of the object and the lighting changes can vary in the area being observed. A part of this report deals with the study on the validity of these modelings with respect to the acquisition configuration: relative locations between the lighting source, the camera and the object, properties of the surface (curvatures and roughness). These models are used to improve feature points tracking in image sequences, by computing simultaneously the photometric and geometric changes. The proposed methods are compared to tracking methods with photometric normalization and the technique proposed by Jin. Both of them compensate for affine photometric changes. Since our approach corrects spatial photometric variations, the robustness and the accuracy of the tracking are improved. Experimental results on specular objects demonstrate the robustness of our approaches to specular highlights and lighting changes, without increasing computation times. These procedures provide a good accuracy of the points location during the sequence.

Mots clés: lighting; specular reflection; photometric models; tracking.; tracking; Illumination changes

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Michèle Gouiffès, Christophe Collewet. Modeling of photometric variations for matching correspondences, its validity and its application to tracking. 2007. ⟨hal-00430799⟩