Efficient Computing for Immersive VR: Real-Time Panoramic Video Stitching on GPU Platforms

The growing demand for immersive Virtual Reality (VR) experiences necessitates seamless stitching of panoramic video. VR unlocks the potential for applications like deep-sea exploration and self-driving car visualizations where a real- time first-person perspective is crucial. Existing panoramic video stitching methods face challenges in achieving real-time performance with multiple videos. Most solutions are offline, limiting their applicability. We explore different approaches to video stitching and propose an optimized stitching pipeline utilizing OpenGL, a widely available graphics library, to leverage the superior processing power of GPUs. By employing off-the-shelf GPUs, we aim to make our solution accessible and cost-effective for a broad range of applications. Post-stitching, a stereo vision module enriches depth perception when using VR headgear. Our approach achieves a significant improvement in stitching speed compared to traditional CPU-based methods. By harnessing the power of OpenGL, we observed a remarkable 17x frame rate increase (reaching 450fps with three video streams). The substantial improvement in frame rate paves the way for a new generation of immersive VR experiences with a seamless, real- time first-person view.