TOP  >  Thesis/Dissertation  >  Shape and BRDF Estimation Utilizing Tyndall Effect with Separation of Scattering and BRDF Parameters

Shape and BRDF Estimation Utilizing Tyndall Effect with Separation of Scattering and BRDF Parameters

Acquisition of shape and reflective characteristics for object modeling is a very challenging research problem that has attracted many researchers. Most of the proposed methods rely on imagery obtained from a camera-light source pair and cameras observe the reflection of the light source from the object surface. The physical model for direct reflection images is quite complex and highly under-constrained; therefore, acquisition of reflective characteristics tends to be ill-posed when only a limited number of source images exist. To overcome this problem, Koyama et al. proposed a novel method utilizing Tyndall effect. Their setup is capable of observing the reflection location as well as the reflection in nearly all possible directions from a single point on an object. Their method uses models for scattering and reflection to simulate the observed images, and non-linear optimization performed to find best-fitting parameters. However, the results show that this method is under-constrained and finds an infinite set of plausible parameters. Moreover, the estimation phase takes more than a day on a super computer. We consider that the source of the problem can be tracked to the simultaneity of scattering and reflection parameters estimation along with the absence of a reference object to create an initial mapping to parameter space from the pixel value space. We propose a new method that separates estimation phases of the scattering and reflection parameters, and introduce a reference object to compare against in order to set the overall brightness of the target objects. Experimental results with several objects shows that our method can stably estimate shape and the scattering parameters and can acquire reflective characteristics at a certain stability. As for computational costs, our method reduced the computation time to the order of seconds in a consumer level desktop computer.