Experiences with Point Cloud Registration

The development of LADAR (laser distance and ranging) technology to acquire 3D spatial data made it possible to create 3D models of complex objects. Because an unobstructed line-of-sight is required to capture a point on an object, an individual LADAR scan may acquire only a partial 3D image, and several scans from different vantage points are needed for complete coverage of the object. As a result there is a need for software which registers various scans to a common coordinate frame. NIST is investigating direct optimization as an approach to numerically registering 3D LADAR data without utilizing fiduciary points or matching features. The primary capability is to register a point cloud to a triangulated surface - a ?TIN? surface. If a point cloud is to be registered against another point cloud, then the first point cloud is meshed in order to create a triangulated surface against which to register the second point cloud. The direct optimization approach to registration depends on the choice of the measure-of-fit to quantify the extent to which the point cloud differs from the surface in areas of overlap. Two such measuresof- fit have been implemented. Data for an experimental evaluation were collected by scanning a box, and registration accuracy was gauged based on comparisons of the volume and height to known values.