High-resolution three-dimensional sensing of fast deforming objects

Abstract

In applications like motion capture, high speed collision testing and robotic manipulation of deformable objects there is a critical need for capturing the 3D geometry of fast moving and/or deforming objects. Although there exists many 3D sensing techniques, most cannot deal with dynamic scenes (e.g., laser scanning). Others, like stereovision, require that object surfaces be appropriately textured. Few, if any, build high-resolution 3D models of dynamic scenes. This paper presents a technique to compute high-resolution range maps from single images of moving and deforming objects. This method is based on observing the deformation of a projected light pattern that combines a set of parallel colored stripes and a perpendicular set of sinusoidal intensity stripes. While the colored stripes allow the sensor to compute absolute depths at coarse resolution, the sinusoidal intensity stripes give dense relative depths. This twofold pattern makes it possible to extract a high-resolution range map from each image in a video sequence. The sensor has been implemented and tested on several deforming objects.