A robust automatic registration method for hand-held structured light 3D scanner

Abstract

There is currently a very strong need for building 3D models of the visible surface for a wide variety of objects in shape and size. A hand-held 3D scanner is a useful tool in many situations, in case of the single measurement range size, the sensor-tracking devices or surface markers are need to realize multiple view alignment, thus limiting their functionality. We propose an efficient fast registration methods based on both texture and geometry which can bring in additional information and compensate for ambiguities in the other cues. Together, we can use rotation-invariant geometric or photometric feature descriptors to extract faithful corresponding points for matching without sensor-tracking devices or surface markers. Meanwhile, range data alignment based on photometric properties is performed better using a RANSAC algorithm to rule out mismatching. Experimental results with real objects indicate the effectiveness of the proposed approach. We have applied this method to hand-held structured light 3D scanner. It realizes a single view 3D measurement within 0.12S and Real-time global registration.