Extending epipolar geometry for real-time structured light illumination

Abstract

Structured light illumination, scanning along both horizontal and vertical directions, achieves more robust accuracy. By introducing the constraint of epipolar geometry, we previously proposed real-time 3D reconstruction using lookup tables; however, we only knew these offline derived tables were the combinations of the elements in calibration matrices of a camera and a projector, and suffered from long-time computation. In this Letter, by parameterizing the line perspectively mapping a 3D world coordinate into the camera and projector spaces, we propose to extend the epipolar analysis by defining phase and optical poles. Thus, we can geometrically address these parameters via analytic closed-form equations, with which we can (1) directly derive lookup tables in real time from the calibration matrices and (2) optimally reduce the number of tables from 11 to 5 to save much more memory space while further accelerating the processing rate. Experiments show that with the same level of accuracy, we significantly reduce the time to compute the lookup tables from more than 20 min to 20 ms, and increase the speed of computing point clouds from approximately 320 to 492 fps.