An illumination-computation acceleration structure based on the sparse voxel directed acyclic graph (SVDAG) is proposed. By merging the same nodes from bottom to top, the sparse voxel octree is converted into a SVDAG, and the polysemy of spatial positions can be eliminated by using the traversal paths and the child masks of the given nodes. Aiming at the closed geometry, an algorithm based on the double depth maps can be used to merge adaptively the nodes located in the closed region, which can further reduce the storage cost while the performance of illumination computation is maintained. A inter-frame multiplex method of SVDAG based on the time correlation is proposed in which all frames of the dynamic scene are used to constitute an integral SVDAG acceleration structure, which can improve the update rate. The experimental results indicate that the rendering efficiency of three-dimensional scene based on the new algorithm is enhanced. When a high resolution dynamic scene is conducted, a relatively high frame rate still can be obtained.