Estimation of light source information from image sequence

Abstract

SUMMARYThis paper proposes an algorithm that analym shading scenes with motion based on a shading model and estimates an incident direction of a light source illuminating an object. If the incident direction of a light source can be estimated. this means that an important environmental parameter of a scene is obtained which enables precise estimation of such parameters as shape and motion in the scene. Computer simulations have been performed for computer-generated images and real images, and it was verified that the algorithm works correctly.

An Illumination Effect Descriptor for Video Sequences

“…This is of high importance for its practical use. In contrast, the extraction of descriptors of static illumination effects or of descriptors of light sources are restricted either to unicolored scene background [7,24], or to unicolored objects [9,21], or they need additional input data as 3D object shape [13,27,29]. The proposed descriptor extraction method shall consider errors in all used data, i.e., the video images and the used motion information.…”

“…The evaluation of temporal changes does not require the mentioned restrictions for objects or background. Also, measuring temporal changes of shading or cast shadows on surfaces of moving objects has been shown to be successful for illumination estimation [13,27,29], for shape estimation [22], for motion estimation [19,30,32], and for shadow detection and tracking [28].…”

Augmented reality with automatic illumination control incorporating ellipsoidal models

“…In this paper, an illumination estimation method allowing for arbitrarily colored natural objects is desired. Such methods from the literature [11], [21] observe the displaced frame difference (DFD), the frame-to-frame difference after motion compensation. The observations may contain stochastic errors such as camera noise.…”

“…In this paper, illumination estimation is based on the approach in [21]. Whereas in [11] only a single point light source is considered, the approach in [21] considers additionally ambient light that is often present in natural scenes. In the following, the illumination estimation method from [21] will be reformulated as maximum-likelihood (ML) estimator.…”

“…First approaches to automatic illumination estimation evaluate only a single video image and assume unicolored shaded objects in the natural image sequence, see, e.g., [16] or [28] for one or [26] for several light sources. More recent approaches evaluate two video images and allow for arbitrarily colored objects, see [11] for a single light source and [21] for a single light source with ambient light. In this paper, the approach of [21] will be chosen because it considers ambient light, too.…”