Point Based Illumination Condition Matching Approach to Photometric Registration

With recent and continued increases in computing power, and advances in the field of computer graphics, realistic augmented reality environments can now offer inexpensive and powerful solutions in a whole range of training, simulation and leisure applications. One key challenge to maintaining convincing augmentation, and therefore user immersion, is ensuring consistent illumination conditions between virtual and real environments, so that objects appear to be lit by the same light sources. This research demonstrates how real world lighting conditions can be determined from the two-dimensional view of the user. Virtual objects can then be illuminated and virtual shadows cast using these conditions. This new technique uses pairs of interest points from real objects and the shadows that they cast, viewed from a binocular perspective, to determine the position of the illuminant. This research has been initially focused on single point light sources in order to show the potential of the technique and has investigated the relationships between the many parameters of the vision system. Optimal conditions have been discovered by mapping the results of experimentally varying parameters such as FoV, camera angle and pose, image resolution, aspect ratio and illuminant distance. The technique is able to provide increased robustness where greater resolution imagery is used. Under optimal conditions it is possible to derive the position of a real world light source with low average error. An investigation of available literature has revealed that other techniques can be inflexible, slow, or disrupt scene realism. This technique is able to locate and track a moving illuminant within an unconstrained, dynamic world without the use of artificial calibration objects that would disrupt scene realism. The technique operates in real-time as the new algorithms are of low computational complexity. This allows high framerates to be maintained within augmented reality applications. Illuminant updates occur several times a second on an average to high end desktop computer. Future work will investigate the automatic identification and selection of pairs of interest points and the exploration of global illuminant conditions. The latter will include an analysis of more complex scenes and the consideration of multiple and varied light sources.