抄録
The state-of-the-art phase-based scanner is suitable for acquiring 3D shapes of large facilities, because it can acquire hundreds of millions point data in several minutes. Since acquired data are very noisy, smoothing operations are very important. However, point data from large facilities are too large to process in the memory of common personal computers. This paper proposes a new streaming smoothing operator for very large-scale point-cloud acquired by phase-based or time-of-flight scanners. Our method reads point data on a hard disk and sequentially processes them in a small region on RAM. This method is based on the fact that point data from a laser scanner can be converted into a height-field on a sphere. Our experimental results show that our method could successfully produce a smooth mesh model from large-scale point data with the limited memory size.
