Point Cloud Compression with Triangular Prediction for Encoding Efficiency

The patent divides the point cloud data into multiple prediction units (e.g., 8x8x8 voxels) and further segments each unit into multiple triangles. This segmentation allows the system to process and encode only the essential geometric information (triangle vertices and normals) rather than all point data, significantly reducing the data quantity while maintaining three-dimensional representation accuracy and improving encoding efficiency.

The patent extracts and utilizes only the necessary geometric information from the point cloud data, specifically the triangle vertices and normal vectors, for prediction and encoding. By taking out only these critical elements and discarding redundant point cloud data, the system achieves efficient encoding while preserving the essential three-dimensional structure needed for VR, AR, and autonomous driving applications.

The patent performs preliminary actions by pre-defining the triangular approximation framework and prediction unit structure before actual encoding. The system pre-establishes the relationship between prediction units, triangles, and motion compensation, allowing the encoding process to follow a standardized procedure that reduces computational complexity while maintaining high encoding efficiency through the pre-planned triangular representation approach.

The patent changes the representation parameters from raw point cloud data to triangular geometric primitives (vertices and normals). This parameter transformation simplifies the data structure and enables efficient inter-frame prediction by working with the simplified triangular representations rather than complex point cloud datasets, thereby reducing processing complexity while improving encoding efficiency.

The patent segments the entire point cloud into multiple prediction units (e.g., 8x8x8 voxels) and applies motion compensation only to each unit independently. This segmentation allows parallel processing of different units, reducing total processing time while maintaining high prediction accuracy within each unit. The triangular approximation within each unit further accelerates computation by reducing the number of points requiring motion compensation.

The patent applies motion compensation selectively to prediction units that require it, rather than processing the entire point cloud uniformly. By applying motion compensation only where necessary (partial action) and using the efficient triangular approximation method, the system achieves high prediction accuracy while minimizing processing time compared to applying the same treatment to all point cloud data.