What Is Point Cloud Compression?

Point cloud compression is the set of techniques and standards for encoding three-dimensional spatial data sets into compact representations that can be stored, transmitted, and decoded while preserving sufficient geometric and attribute fidelity for downstream applications. A point cloud is a collection of individual points in 3D space, each defined by XYZ coordinates and typically carrying color, reflectance, or other per-point attributes captured by LiDAR sensors, structured-light scanners, or photogrammetric reconstruction systems. Uncompressed point clouds can easily reach hundreds of megabytes per frame or gigabytes per scene, making compression essential for real-time streaming, network transmission, and storage on resource-constrained devices. The field draws from signal processing, computer graphics, information theory, and the video coding community, and it has accelerated significantly since the Moving Picture Experts Group (MPEG) launched a formal standardization effort in 2017.

Geometry Compression

The geometric component of a point cloud encodes the 3D positions of millions of individual points. A foundational structure for geometry compression is the octree, a hierarchical spatial subdivision in which a bounding cube is recursively divided into eight sub-cubes, with occupied cells encoded as binary flags. Octree-based coding achieves compact representations for sparse point clouds by skipping empty regions entirely. An alternative is the triangular prism surface approximation used in trisoup coding, which fits local surface patches to groups of points and encodes them as a mesh fragment, reducing the per-point storage requirement for smooth surfaces. Deep learning approaches using 3D convolutional networks and sparse convolution architectures have more recently been applied to geometry coding, with learned entropy models producing competitive rate-distortion results. The geometry coding pipeline typically quantizes point positions to a voxel grid before entropy coding, and the choice of voxel resolution directly controls the trade-off between spatial fidelity and bits consumed.

Attribute and Color Compression

Beyond geometry, point clouds carry per-point attribute data such as RGB color, normal vectors, and reflectance intensity. Attribute compression builds on the geometry representation to predict attribute values from neighboring points, reducing the residual that must be entropy coded. The Attribute Transform Scheme (ATS) used in MPEG's Geometry-based Point Cloud Compression (G-PCC) standard applies a lifting-based wavelet transform over a nearest-neighbor graph built from the decoded geometry, compressing smooth color variations efficiently while preserving sharp boundaries. Region Adaptive Hierarchical Transform (RAHT), another G-PCC attribute coder, processes the octree level by level using Haar-like transforms, achieving strong performance on sparse LiDAR data. As reviewed in the IEEE Journal on Emerging MPEG Standards for Point Cloud Compression, the coupling of geometry and attribute coding pipelines is one of the key design decisions distinguishing the two MPEG codec families.

Standards: V-PCC and G-PCC

MPEG developed two parallel standards to address the different characteristics of dense and sparse point clouds. Video-based Point Cloud Compression (V-PCC) projects the 3D surface of a dense, object-centered point cloud onto a 2D image atlas using a patch-based mapping, then applies conventional HEVC or VVC video codecs to compress the resulting image sequences. This approach delivers high compression efficiency for human-body captures and immersive video content by reusing decades of video coding infrastructure. Geometry-based Point Cloud Compression (G-PCC) operates directly in 3D, making it well suited to sparse outdoor LiDAR scans where the 2D projection approach leaves large empty regions. As the ACM GetMobile introduction to point cloud compression standards explains, both standards target specific use cases rather than providing a universal solution, and selecting between them depends on the density, motion characteristics, and delivery requirements of the point cloud data. The MPEG Point Cloud Compression project site maintains the reference software and conformance data sets for both standards.

Applications

Point cloud compression has applications in a range of fields, including:

• Autonomous vehicles and robotics, where LiDAR-generated point clouds must be processed and shared in real time
• Augmented reality and virtual reality, including immersive 3D displays of real-world scenes and volumetric video
• Cultural heritage and architecture, for compressed archival of high-resolution 3D scans of historical sites
• Geographic information systems (GIS) and smart city mapping using airborne LiDAR surveys
• Industrial inspection and quality control, transmitting 3D scan data from factory floors to remote analysis systems