Real-Time Dense 3D Reconstruction with Active Vision System Via Graphic Hardware Acceleration

摘要

We herein present a real-time active 3D imaging system. The system consists of a pair of stereo camera and a projector. We develop two sets of algorithm using different configurations on our system. The first uses a stereo camera,by projecting fringe patterns onto target surface,dense disparity map can be computed in sub-pixel accuracy even for region without any feature. While the second uses only one of the cameras and the projector is used as an inverse camera. The disparity map is computed by unwrapping the wrapped phase map which is derived from three-phase-shifting method, then triangulation is used to computed 3D information. Graphic hardware is used to accelerate the computation. Experiments show that both methods exhibit good performance in accuracy,density and speed,can capture 3D image at the speed of 10 frames per second with a maximum size of 300k points per frame.