The segmentation of a 3D object into parts has customarily been determined by analyzing the object’s geometric properties, such as surface curvature or volumetric shape. This paper uses a novel physics-based approach on range images to segment a 3D object into parts. A charge density distribution is simulated over a surface that has been tessellated by a triangular mesh. The deep surface concavities are detected first, using the physical fact that electrical charge on the surface of a conductor tends to accumulate at a sharp convexity and vanish at a sharp concavity. Deep surface concavities are found by tracing the local charge density minima and then decomposing the object into parts at these joints. The computational complexity is O ( N2 ) , where N is the number of triangular facets. The approach does not require an assumption of surface smoothness; it uses weighted global data to produce robust local surface features for segmentation. The method produces a unique description of the surface, and the segmentation is invariant to object scale, translation, and rotation. However, the technique may fail for certain shape configurations containing holes.
Experimental results of segmentations of several objects are given. The actual computing time for one of the objects (an owl) on an SGI R8000 workstation was 90 seconds for the charge distribution, with about two seconds for surface triangulation, and less than a second for part decomposition.
The paper is clearly written and contains many references to related work.