ABSTRACT
The process of obtaining a 3-D model with real texture based on a series of 2-D
photos taken from different angles around the target is called 3-D surface
reconstruction, which as a valuable technology can be applied to virtual tour, electronic
commerce, etc. The aim of this study is to develop a measuring system of people’s
heads for Hairde Co.ltd. China using 3-D surface reconstruction technology to help the
company with customizing its bald clients’ hairpiece individually.
In this paper, a reconstructed 3-D model based on real 2-D photos collecting,
intersection of all view cones and 3-D surface reconstruction method is obtained. The
process is demonstrated as the following six sections.
PHOTO COLLECTION: In this section, a method of image collection has been
designed, based on which a series of more than 20 photos is collected.
CAPTURING IMAGES: Designing a method of capturing images, according to
which we take more than 20 images.
IMAGES SEGEMENTATION: In this section, the algorithm based on over-
segmentation perceptual grouping is applied, which makes it possible to segment target
areas efficiently and accurately. A newly defined distance is used in the process of edge
detecting ,which makes discontinues border able to be connected to edge. Target
selection is completed by manual, which is fast and simple. To ensure the correctness of
merging, the local merger strategy and global merger strategy were used while regions
merger.
CAMERA CALIBRATION: In this part, the technology of Structure From Motion
(SFM) is applied to realize camera calibration.
3-D SURFACE RECONSTRUCTION: In this section, we discussed and realized a
visual hull reconstruction algorithm which bases on intersection of all view cones in the
surface of the object. Input silhouettes of imaging of the object in different perspective
and the camera projection matrix of these images, then output the object’s 3 d visual
appearance surface model.
TEXTURE MAPPING: An automatic texture mapping method based on projection
is applied in this part which is new and fast. To compute a texture map for each planar
segment, we back project the selected views on to different texture images, and decide
the vertex color according to the relationship of the angle between vertex normal and
II