Verification of Sky Models for Image Calibration

Rishi Ramakrishnan, Juan Nieto, Steve Scheding; Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops, 2013, pp. 907-914


Perception systems operating in outdoor scenarios face challenges due to the high dynamic range of the image, as different regions are illuminated by varying amounts of sunlight and skylight. A pre-processing step like image calibration can be used to convert the pixel values to an illumination independent domain such as reflectance. Each pixel is therefore represented by a characteristic material description, instead of an illumination and viewpoint dependent pixel colour. This assists object identification, segmentation and classification algorithms [20]. This paper investigates modelling the sky colour through a number of parametric approaches typically used in the computer graphics community for rendering purposes, namely those developed by Preetham et al. [17] and HosekWilkie [6]. The models are compared in terms of chromaticity with observations taken from a camera and are used to develop an environment map for the application of inverse reflectometry of diffuse objects in an outdoor environment. This is of particular importance for applications involving imaging objects whose primary illumination source is skylight. It was found that the Hosek-Wilkie [6] model produced more robust estimations and was less sensitive to changes in azimuth, while both models had similar reconstruction results with angular errors of approximately 0.15 radians.

Related Material

author = {Rishi Ramakrishnan and Juan Nieto and Steve Scheding},
title = {Verification of Sky Models for Image Calibration},
booktitle = {Proceedings of the IEEE International Conference on Computer Vision (ICCV) Workshops},
month = {June},
year = {2013}