Depth map estimation with unknown fixed pattern projection

Abstract

Depth map estimation is important in many computer vision applications. Various systems and algorithms have been proposed for depth map estimation. In this paper, we focus on a fixed random dot projection approach. An invisible near infrared (NIR) is usually used for the random dot projection. In many cases, the depth map estimation system of the fixed random dot projection approach consists of an NIR projector, an NIR camera and an RGB camera. First, we propose a simple calibration for the fixed NIR random dot projection and the NIR camera. Then, the depth map estimation algorithm is described. We also propose a depth map refinement algorithm using the RGB camera image. Experimental results demonstrate that the simple calibration is good enough to estimate the depth map. It is also shown that the depth map can be refined by the RGB camera image.