Implicit Surface-Based Geometric Fusion

3-D studio production of animated actor models

“…Displacement map representation of individual stereo shape estimates allows the correspondence between multiple noisy sets of overlapping surface measurements to be established. This overcomes limitations of previous implicit surface [5,7] and mesh-based nearest-point [23,29] range image integration algorithms in the presence of significant measurement noise.…”

“…Advances in active sensor technology together with research in computer vision and graphics has resulted in systems for capturing surface models of complex 3D objects, people and internal environments [5,7,16]. These approaches capture accurate and realistic 3D models of complete objects with a level-of-detail not possible with previous manual techniques.…”

“…Previous algorithms for integration of multiple view range images [5,7,21] assume relatively accurate surface measurements from active 3D sensors using projected structured light. Stereo estimates of surface geometry for people in normal clothing are relatively noisy due to the absence of local variations in surface appearance required for accurate correspondence localisation.…”

“…A high-resolution triangulated mesh M H ij is then constructed by constrained triangulation [7] of adjacent surface position estimates x(r, s), x(r + 1, s), x(r, s + 1), x(r + 1, s + 1). A triangle is added to the mesh if the Euclidean distance between adjacent surface measurements |x(r, s) − x(p, q)| < d t , where d t = 3δx and δx is the surface sampling resolution.…”

“…For a capture volume of 2m 3 with image resolution 720 × 576 we have δx ≈ 0.01m which is used throughout this work. Constrained triangulation has been used in previous range image integration algorithms to construct a mesh which approximates the local surface topology [5,7]. This results in a mesh M H ij which approximates the local surface geometry but does not connect across step discontinuities which occur at occlusion boundaries.…”