Global camera path optimization

A method of three-dimensional reconstruction comprising:

• acquiring a plurality of frames of image data of a surface of an object, each one of the plurality of frames of image data captured from a camera position along a camera path and each one of the plurality of frames of image data providing an independent three-dimensional data set including a conventional image of the object from the camera position and disparity data stored in the frame for a three-dimensional reconstruction of the surface of the object as viewed from the camera position;
• selecting a subset of the plurality of frames of image data to provide a plurality of key frames, each one of the plurality of key frames related to at least one other one of the plurality of key frames by a portion of the camera path including a rotation and a translation determined based upon one or more common points in the three-dimensional reconstruction of the surface of the object in each of the respective key frames, wherein the plurality of key frames provide coverage for a substantial portion of the surface of the object and wherein the remaining ones of the plurality of frames of image data are non-key frames;
• estimating a camera motion between two adjacent key frames wherein the estimating is based on the rotation and the translation;
• optimizing the estimation of the camera motion between the two adjacent key frames by creating consistency among motion parameters using an overdetermined system of motion constraint equations, wherein the motion parameters are comprised of information on the rotation and the translation;
• providing a three-dimensional model of the object based upon the camera path for the plurality of key frames;
• determining a second rotation and a second translation from one of the key frames to at least one of the non-key frames sequentially positioned between the one of the key frames and a sequentially adjacent one of the key frames;
• obtaining three-dimensional reconstruction information of the surface of the object from the camera position of the at least one of the non-key frames to provide upsampled three-dimensional data;
• and
• adding the upsampled three-dimensional data to the three-dimensional model based upon the second rotation and the second translation.