high performance computing on graphics processing units: hgpu.org

Time-of-Flight (ToF) cameras gained a lot of scientific attention and became a vivid field of research in the last years. A still remaining problem of ToF cameras are motion artifacts in dynamic scenes. This paper presents a new preprocessing method for a fast motion artifact compensation. We introduce a ow like algorithm that supports motion estimation, search field reduction and motion field optimization. The main focus lies on real-time processing capabilities. The approach is extensively tested and compared against other motion compensation techniques. For the evaluation, we use quantitative (ground-truth data, statistic error comparison) and qualitative (real environments, visual comparison) test methods. We show, that our proposed algorithm runs in real-time within a GPU based processing hardware (using NVIDIA Cuda) and corrects motion artifacts in a reliable way.