Research Of Real-time Imaging Simulation Technology Based On CUDA

Ray tracing algorithm is used in graphic-based imaging simulation technology to provide simulation images with excellent fidelity. With the help of those images, researchers can predict and estimate optical imaging device's target acquisition ability in various application conditions. Those images can also be used as test data in the area of Image quality evaluation, Image processing algorithm verification and target recognition training. Although the ray tracing algorithm can provide high quality simulation images, due to large amount of computation, it cannot achieve real-time performance in CPU realization.In this paper, GPU's characteristic of highly parallelization is used to accelerate ray tracing process. According to the advantages and limitations of CUDA, this paper did some adaptable modification and realized a ray tracing algorithm framework under the SIMT parallel architecture.In order to further enhance the scene rendering speed, the paper adopts KD-Tree technology to manage the scene elements. Five popular KD-Tree traversal algorithms that can be realized in GPU architecture are compared to find out the most efficient algorithm. At last KD-Tree traversal algorithm with ropes is selected to further reduce execution time.This paper also selected the 3DS Model to organize the scene elements, and a special function was designed to re-organize the scene elements into the format that suitable for GPU.Finally, this paper designs a generic, distensible ray tracing system with modular approach based on ideas of hierarchical organization and provides a basis framework for future research of real-time imaging simulation technology based on CUDA.