Point Cloud Tree Compression With Azimuthal Prediction Contexts

The patent applies preliminary action by performing azimuthal prediction using previously decoded nodes before encoding the current node. The encoder determines azimuthal prediction angles from closest previously decoded nodes and uses these to predict occupancy of child nodes, avoiding the need to recursively split and examine all sub-volumes. This preliminary prediction step reduces computational burden while maintaining compression effectiveness.

The patent implements self-service through inferred direct coding mode where isolated points are automatically identified and encoded without requiring manual intervention or complex recursive processing. The system uses azimuthal prediction to self-identify isolated points based on the absence of occupancy in predicted child nodes, enabling automatic compression of sparsely populated regions with reduced computational effort.

The patent extracts isolated points from the traditional tree-based encoding process by using azimuthal prediction to identify them separately. Instead of signaling all nodes through recursive splitting, the method extracts and specially handles isolated points using inferred direct coding mode, where occupancy is inferred from azimuthal prediction results. This extraction approach reduces the number of signals required and improves bandwidth efficiency.

The patent changes the encoding parameter approach by introducing azimuthal prediction angles and using them to determine occupancy of child nodes. Instead of using fixed recursive splitting parameters, the system dynamically adjusts encoding based on azimuthal prediction results, changing from a uniform tree-based approach to a directionality-aware approach that reduces signaling overhead for isolated points.

The patent applies preliminary action by performing azimuthal prediction before full recursive encoding. The encoder uses azimuthal prediction angles from previously decoded nodes to predict which child nodes will be occupied, allowing it to skip recursive splitting of obviously empty nodes. This preliminary prediction maintains encoding accuracy for occupied regions while reducing processing time by avoiding unnecessary recursion into empty sub-volumes.

The patent implements partial action by performing recursive splitting only for nodes that are predicted to be occupied based on azimuthal prediction. Instead of recursively splitting all sub-volumes to ensure complete encoding accuracy, the method applies partial recursion only where needed, maintaining sufficient encoding accuracy while significantly reducing processing time for sparsely populated point clouds.