Structured light 3D measurement of reective objects using multiple DMD projectors

摘要

Many applications of 3D imaging must deal with objects that are highly specularly reective and have multipleinterreections, resulting in missing and erroneous points. We use a single camera with multiple DMD projectors,resulting in multiple incident lighting angles to deal with both specular and interreections. At each camera pixel,multiple 3D point estimates are computed from each of the camera-projector pairs. These multiple estimatesare fused together with an expectation maximization (EM) clustering algorithm and the maximum likelihoodestimate for each point is obtained. A novel noise filtering algorithm is also presented which makes use of the EMalgorithm covariance to suppress spurious points caused by reections. Results show that our multi-projectorapproach and EM estimator increase 3D measurement accuracy and recoverable points for complex reectiveobjects, while reducing error.