Real-time feature-based image morphing for memory-efficient impostor rendering and animation on GPU

摘要: Real-time rendering of large animated crowds consisting thousands virtual humans is important for several applications including simulations, games, and interactive walkthroughs but cannot be performed using complex polygonal models at frame rates. For that reason, methods numbers precomputed image-based representations, called impostors, have been proposed. These take advantage existing programmable graphics hardware to compensate computational expense while maintaining visual fidelity. Thanks these methods, the number different rendered in real time no longer restricted by power texture memory consumed variety discretization their animations. This work proposes a resource-efficient impostor methodology employs image morphing techniques reduce consumption preserving perceptual quality, thus allowing higher diversity or resolution crowds. Results experiments indicated proposed method, comparison with conventional techniques, can obtain 38 % smoother animations 87 better appearance quality reducing key-frames required animation via resynthesizing them up 92 similarity on time.