A Coded Illumination Pattern Based on PN-sequences for 3-D Robot Vision

Abstract

Abstract A novel illumination coding scheme for use in 3-D image acquisition is presented. It provides a solution to the identification problem encountered in multiple stripe triangulation. A single camera picture yields sufficient information to compute the range image, which makes it possible to deal with moving scenes. The projected pattern consists of a spatial arrangement of easily detectable features each of which indicates a single bit at the corresponding sample point. The bit map is such that every local neighbourhood of some minimal size exhibits a unique bit pattern. This means that individual strikes can be identified merely by their local signature in the camera image. The bit map is further optimized to enhance fault tolerance by maximizing the discrimination between the local bit patterns in terms of Hamming distance. A superposition of pseudo-noise sequences meets these requirements. An optimal bit pattern has been determined for a lattice of 64x63 sample points. The experimental setup consists of a slide-projector and a solid-state camera. A program package includes routines for locating the sample points in the image, recovery of their signature, triangulation and system calibration.