Multi-view 4D Scene Reconstruction through Structured Light Methods
Ricardo Garcia and Avideh Zakhor
In computer vision, significant research efforts have gone towards developing accurate 3D models of objects or scenes through the use of depth reconstruction methods such as stereo vision and structured light. Structured light systems are similar to stereo vision except, rather than using a pair of cameras to capture depth, one of the cameras is replaced by a projector. This projector illuminates the scene with unique position based codes that are captured by the remaining camera and used to triangulate depth. Using both stereo cameras and structured light techniques, researchers have been able to extend 3D modeling methods to capture 3D models of time-varying objects. These 4D captures, where we consider time as the fourth dimension, have mostly focused on recreating a model from a single perspective of the dynamic scene. This is especially true in techniques that utilize structured light methods.
In this work, we are interested in developing a structured light system that is able to capture a dynamic scene from multiple perspectives in an effort to build a complete model that can be viewed from any direction. Because structured light adds information to the scene to determine the 3D structure of an object, special considerations must be made to limit the interference between projectors as they illuminate the scene from different directions. Technical challenges in this work will also include calibration of multiple cameras and projector, synchronization among multiple cameras and projectors, and merging single view reconstructions into a single model. Our final system should be able to capture a moving scene contained within a volume that is at least 1 x 1 x 1 m in size.