MSU: 3D graphics optimization method using point clouds

2025: Introduction of 3D graphics optimization method using point clouds

Scientists at Moscow State University have proposed a method for optimizing 3D graphics using point clouds. The university announced this on April 21, 2025.

VMK MSU researchers have developed a method for automatically creating point clouds with levels of detail to optimize the rendering of high-polygonal models. The approach significantly improves the performance of graphics applications without losing image quality.

Modern 3D graphics models are becoming more detailed, with millions of polygons and high-quality textures. This allows you to create photorealistic graphics in video games virtual reality and scientific simulations, but at the same time leads to high demands on computing resources. A large number of polygons load graphics (processor GPU), reducing frame rate and slowing down applications.

Scientists at VMK MSU have proposed a method that allows you to optimize the rendering of such models using point clouds with dynamic levels of detail. Unlike traditional polygonal grids, point clouds are a set of individual points evenly distributed over the surface of an object. This approach allows you to use less data to display models, while maintaining the visual quality of the image.

The method is based on converting complex polygon models into point clouds with different densities. This is achieved by using an algorithm for distributing points over blue noise, which allows you to evenly cover the surface of the model, creating a visual illusion of continuity. An important feature of the method is the dynamic change in the density of point clouds depending on the distance to the camera and the viewing angle. This means that the farther the object from the camera, the fewer dots are used to display it, and the density increases as the camera approaches.

The process begins by generating a cloud of points that are evenly distributed across the model surface. This provides a visual match to the original geometry from any viewpoint. This uses the Poisson Disk method to filter points and create levels of detail. This avoids dots in one place and creates a more realistic image.

Then there is a dynamic change in the density of the point cloud. The nested structure of the cloud allows you to change the number of rendered points in real time depending on the distance to the camera. This is achieved by calculating the required density based on the camera parameters and the position of the object. For example, if an object is far away, it is displayed with a lower point density, and when approached, the number of points automatically increases. This allows optimal use of computing resources without overloading the GPU.

The final step is to render the point cloud. For this, conservative rasterization is used, which allows you to increase the screen coverage and avoid the appearance of artifacts. This is especially important to prevent breaks and holes in the image when drawing objects from different angles. Optimizations such as trimming invisible points to normal and selecting the level of detail depending on the position of the object on the screen are also used. This allows you to achieve high performance without losing image quality.

The developed method has several key advantages compared to traditional rendering methods. First, it significantly reduces the amount of data required to display objects, which reduces the load on the GPU and speeds up applications. Second, the dynamic change in dot density provides high performance without losing image quality, since the object is always displayed with optimal detail depending on the camera position.

This method can be useful in various graphic applications such as video games, virtual and augmented reality systems, as well as visualization in scientific and engineering calculations. For example, in virtual reality, it allows you to optimize the display of complex scenes with high detail, while maintaining a high frame rate. In scientific simulations, the method helps to visualize large amounts of data, such as the results of numerical modeling, with high speed and accuracy.

Our method allows you to effectively use point clouds to render complex 3D models. This is especially important in graphics applications that require high performance and image quality. We have shown that with the help of dynamic levels of detail, impressive results can be achieved without compromises in visual quality, - said Alexander Shcherbakov, junior researcher at the Laboratory of Computer Graphics and Multimedia, VMK Moscow State University.

Future research will focus on integrating the method with new graphical APIs and applying to other types of geometry, such as voxel models and laser scanning data. It is also planned to adapt the method for use in virtual reality systems, taking into account the peculiarities of displaying on VR helmets, such as viewing angle and screen resolution.