CLOSE-RANGE OBJECT 3D SURFACE MODELING USING MULTI-CAMERA AND MULTI-IMAGE MATCHING

摘要

The generation of photo-realistic 3D model of close-range objects is an important task for cultural heritage digital documentation. In recent years, the laser scanning has been widely used due to high level of detail and high degree of accuracy can be achieved. However, the cost is high and the data acquisition and post-processing are still labor intensive. On the other hand, the image-based 3D modeling has the advantages in many aspects, not only its fast data acquisition but also the texture and embedded semantic information can be extracted from the image. However, the lack of efficiency in photo-triangulation is unavoidable unless an accurate and reliable automatic tie-point matching tool is available and a strong imaging geometry could be established. This paper thus proposes another image-based 3D modeling scheme using multi-camera configuration and multi-image matching technique. There is no need for human intervention during photo-triangulation. Five digital cameras were used and fixed on a mental bar. The interior and relative orientation parameters of all cameras can be calibrated automatically using Australis© before or after image acquisition. These orientation parameters can then be applied to all photos that were taken using the same multi-camera configuration. That means, when performing multi-image matching using CLORAMA© for surface point cloud generation, these orientation parameters will remain at the same values. In order to complete the modeling of a large piece of object, multiple stations are required and surface stitching is necessary. Several experiments were conducted using the proposed scheme. In which, one case was compared with a reference model acquired by ATOS-I 2M and results in a 0.31mm of absolute accuracy. It demonstrates that the proposed scheme is accurate, reliable, inexpensive, fast data acquisition and without the need of photo-triangulation.