Approximate Regularization for Structural Optical Flow Estimation

Range images registration by correction of perspective deformations and camera calibration

“…We do not incorporate additional constraints which control the parameters of matrix K A from Equation (5). Instead, we use the appropriate weights ω i .…”

“…Equations (5) show that the new camera matrix is computed by the correcting the old camera matrix by the upper triangular matrix K A . We consider this process as the camera calibration correction.…”

“…Our strategy includes the asynchronous capturing of the object from different perspectives (2-3 images from 3-6 angles). Surface parts are then reconstructed separately by using the so called structural optical flow algorithm [5]. The reconstructed parts are matched together.…”

“…Modern methods for estimating 3D models usually consist of three steps: reconstruction of surface parts, matching and stitching. The first step can be done by different techniques such as laser scanners [1], structured light projection [2], time of flight (TOF) cameras [3] or multi-frame reconstruction [4,5]. Matching is necessary in order to position all surface parts in the common coordinate system.…”