In this paper, we present a prototype system, which uses CAVE-based virtual reality to enhance immersion in an augmented reality environment. The system integrates virtual objects into a real scene captured by a set of stereo remote cameras. We also present a graphic processing unit (GPU)-based method for computing occlusion between real and virtual objects in real time. The method uses information from the captured stereo images to determine depth of objects in the real scene. Results and performance comparisons show that the GPU-based method is much faster than prior CPU-based methods.