high performance computing on graphics processing units: hgpu.org

Recent methods in augmented reality allow simulating mutual light interactions between real and virtual objects. These methods are able to embed virtual objects in a more sophisticated way than previous methods. However, their main drawback is that they need a virtual representation of the real scene to be augmented in the form of geometry and material properties. In the past, this representation had to be modeled in advance, which is very time consuming and only allows for static scenes. We propose a method that reconstructs the surrounding environment and estimates its Bidirectional Reflectance Distribution Function (BRDF) properties at runtime without any preprocessing. By using the Microsoft Kinect sensor and an optimized hybrid CPU & GPU-based BRDF estimation method, we are able to achieve interactive frame rates. The proposed method was integrated into a differential instant radiosity rendering system to demonstrate its feasibility.