Ray tracing method based on GPU

Abstract

The invention provides a ray tracing method based on GPU. The method comprises the following steps: a CPU side loads a scene model, acquires triangular plate metainformation, texture information and light source information by analysis, and packaging and sending the information to a GPU side; the GPU side acquires the triangular plate metainformation, the texture information and the light source information, and respectively stores the information to an overall storage and a constant storage correspondingly; the GPU side builds a Kd-Tree using the scene model of a whole 3D scene as a root node; the GPU side performs ray tracing operation; the CPU side performs post image processing; and the CPU side reads pixel information in an image cache area, and displays the information to the screen to complete rendering. According to the method provided by the invention, the ray tracing efficiency is improved by reasonably calling the CPU and the GPU to work cooperatively. Compared with the traditional method adopting the CPU to render, the method provided by the invention is significantly improved in rendering efficiency, and can complete the model rendering work with the same effect in shorter time.

Description

technical field [0001] The invention belongs to the field of computer graphics, in particular to a GPU-based ray tracing method. Background technique [0002] The ray tracing method is used to generate a realistic three-dimensional virtual scene, which is one of the cores of computer graphics. Since the day it was proposed, it has been widely used in film and television production, 3D simulation, computer-aided design and other fields, and has greatly promoted the progress of virtual reality technology. This method is a general technology based on geometric optics. It traces the light that interacts with the object to obtain the path of the reflection and scattering of the light on the surface of the object, which is used to simulate and generate a real virtual scene. It is considered It is the future direction of graphics processing. The ray tracing method needs to trace each ray in the scene, which involves the refraction and reflection of each ray, and finally completes...

AI Technical Summary

Problems solved by technology

The literature "Fast KD-Tree Construction Method for Heuristically Exploring the Optimal Segmentation Plane" proposes a fast construction method based on the binning algorithm. This method first analyzes the cost function of the kd-tree and heuristically locates the The sub-interval where the segmentation plane of the current node is located; secondly, further refinement sampling (sub-sampling) is performed on the detected sub-interval, so that the obtained segmentation plane is better close to the optimal segmentation position, but the article implements The single-threaded version of the algorithm does not optimize memory management, does not fully utilize the advantages of computer parallel processing, and there is still a lot of room for improvement

The literature "Physically-Based Distributed Parallel Ray Tracing Algorithm" proposes a dynamic adaptive distributed parallel ray tracing method for Stanford University's classic multi-threaded ray tracing engine - PBRT. The task method provides a coarse-grained rendering unit for distributed parallel ray tracing in computer clusters; then, after analyzing and comparing the advantages and disadvantages of static and dynamic task scheduling strategies in the distributed ray tracing method, a static and dynamic task allocation is proposed This strategy improves the load balance between nodes. This method has good efficiency and scalability, and improves the parallelism of the ray tracing method. However, the algorithm only uses CPU clusters and does not consider the use of GPUs to improve rendering efficiency.

[0004] It can be seen that although the above methods have greatly improved the efficiency of traditional ray tracing, under the current computer architecture, the performance of the computer cannot be fully utilized, and there is a problem of low rendering efficiency in the actual application process, which cannot satisfy actual needs

Images