Simulating global illumination has received significant attention in computer graphics research for many years. Many effects provide important visual cues that help viewers understand spatial relationships, and drastically improve the realism of a rendered scene, such as diffuse inter-reflection, caustics, refractions and soft shadows. For many real-time applications, computing global illumination in real time is significant importance. This paper proposes an algorithm combining reflective shadow maps, screen space ambient occlusion and percentage closer filtering for realistic image synthesis. Our method changes sampling patterns and accumulates the results of previous frames to improve the quality. In order to achieve high performance, this paper modified the reverse re-projection technique to converge information during camera movement. Finally, the proposed method is demonstrated with various experimental results. The performance is also shown to be good for large-scale scenes at interactive frame rates.