Fusion of stereo image data from multiple views for 3D data acquisition

Abstract

The 3-dimensional reconstruction of the surface of an object has been studied extensively over the years, mainly due to its applications ranging from industrial product inspection, to robot guidance, and 3-dimensional object modelling. The remaining open issues, however, include the acquisition of accurate depth information and the presence of occlusions in the images. This research concerns the acquisition of 3-dimensional data from images for the specific purpose of modelling a person's head. A multiple stereo camera vision platform was designed to alleviate the aforementioned open issues, and various techniques which can be used to fuse together the visual information acquired by the multiple cameras were also investigated. Passive and active lighting techniques were used to address the lack of distinguishable texture on the human face. Furthermore, techniques which permit an improvement in the reconstruction resolution were studied. The results obtained for the multi-camera setup revealed an increase in both the precision of the computed depth information and the robustness of the system to occlusions, over the results obtained for the single-stereo setup. In addition, the combination of passive and active lighting techniques reduced the number of false matches in the 3-dimensional reconstruction, particularly in smooth and textureless regions on the human face. Furthermore, the implemented super-resolution technique provided an inexpensive solution to the resolution enhancement problem. The results which were obtained from the proposed implementation were very promising.