3-D reconstruction engine
View Patent ↗Systems and methods for generating three-dimensional models of an object use images having unmeasured camera parameters. Camera calibration determines the perspective of the camera from the content of the images. A background having a pattern with a known marks in each image can facilitate determination of the camera parameters. One background pattern includes separated marks having rectangular sections where corners of the rectangular sections provide calibrations points for the camera parameters. The camera parameters can also be determined by matching features of the object in different images and determining differences in perspective from differences in the appearance of the matched features in different images. A combination of projective and metric reconstructions provides robust reconstruction.
1. An image processing method comprising:
(a) analyzing a set of images containing an object and a background to identify for each image a silhouette of an object and locations in the image of pattern points in the background;
(b) for each image, using the locations of the pattern points to determine a transform from three-dimensional coordinates to two-dimensional coordinates in the image;
(c) transforming three-dimensional coordinates of a set of points in a candidate volume to the two-dimensional coordinates of a first of the images;
(d) identifying an approximate volume of the object as containing the points that the transform maps onto the silhouette of the object in the first of the images;
(e) setting a current candidate volume equal to the approximate volume;
(f) transforming three-dimensional coordinates of a set of points in the current candidate volume to the two-dimensional coordinates in a next of the images; and
(g) identifying the approximate volume of the object as a minimal volume containing the points that transform onto the silhouette of the object in the next of the images.
2. The method of claim 1 , further comprising repeating the steps (e), (f), and (g) of claim 1 one or more times.
3. The method of claim 1 , further comprising using the points in the approximate volume to construct a mesh of polygons that collectively approximate the surface of the object.
4. The method of claim 3 , further comprising mapping a texture from the images to the polygons.