A least-squares technique for shape recognition with depth data

Abstract

The use of spherical harmonic coefficients to describe objects represented by depth data is introduced. In a previous implementation, it was necessary to fit planar patches to the data before coefficient calculation, and numerical integration techniques were used in the computations. Here a least-squares implementation is presented that allows for coefficient calculation directly from the depth data without the use of planar patches or numerical integration. This technique is much faster, and experimental results are shown to demonstrate its effectiveness.<>