Structured lighting 3D reconstruction and 3D shape matching of human model for garment industries

Abstract

This dissertation presents an overall methodology for structured lighting 3D reconstruction and new ideas in 3D shape matching of human model for garment industries. With the suggested methodology in 3D reconstruction, a low cost, efficient and accurate 3D multi-camera human body scanning system can be established. In addition, the 3D shape matching methods investigated in this thesis also shows a great potential in fulfilling both global and local features matching. The first section, calibration of multi-camera system, is the initial task for building up a scanning system. Simple plastic foam made calibration target is used. Unlike self-calibration method, with a single target, a single global world coordination system is introduced to all cameras observing the calibration target. Also, as we are working on the structured lighting 3D reconstruction, calibration of the projected structured light patterns uses the same target. The whole process is efficient and accurate. The resulting 3D point cloud reconstruction can be easily accomplished within the required degree of accuracy. The second part deals with the mesh model reconstruction of human model from sparse point cloud data and data with occluded region. A base-plane method is adopted to construct the rough model and the model is refined with selected radial basis function (RBF) augmented by a template model to achieve the goal. Finally, skeleton based 3D shape matching and with Laplacian Eigenmap 3D shape matching methods are investigated. Both techniques are studied for extracting the local features on a human model. Clear potential for global and local shape matching using a combination of the two methods has been observed. With those techniques, constructing an accurate 3D human body model using a low cost 3D multi-camera scanning system is possible.