Abstract: Image-based reconstruction is an important researching content of computer graphics and virtual reality. However, current methods are still difficult to achieve fast 3D reconstruction of real objects to support virtual-reality applications. This paper presents a novel method to implement real-time 3D reconstruction and virtual-reality based on GPU. First, we split the entire space of real objects into separate voxels, and checked the state of each voxel by the combination of Visual Hull and Exact Marching Cube. Second, we compressed voxel state sequence and triangularize the non-empty voxels collection in parallel, and then implemented pixel-based texture mapping by using multiple texture mapping and programmable functionality of graphics hardware. Third, the particles of virtual objects are assumed to be the motion-confined fluid particles, so the virtual-reality interactions are computed by solving the fluid motion equation using parallel smoothed particle hydrodynamics. The results demonstrate that the proposed method can achieve real-time 3D reconstruction and virtual-reality interaction in 20 FPS under the space resolution of 32×32×32, and satisfy the requirements of stability and efficiency in virtual reality and computer graphics.