Point Cloud Prediction Trees Using Azimuth-Based Branch Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Sub-optimal construction of prediction trees in predictive geometry coding for point clouds leads to coding inefficiencies, particularly when LIDAR systems do not rotate a full 360 degrees, resulting in inaccurate point predictions across different sweeps.
Innovation Solution
Generate prediction trees by determining azimuth differences between points and terminating branches based on azimuth thresholds to connect tree branches effectively, optimizing coding by leveraging inherent dependencies within point cloud data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If prediction trees are constructed without considering azimuth differences between points, then the coding process is simpler, but coding efficiency deteriorates due to inaccurate predictions
Solution Approach 1:
The prediction tree is segmented into multiple branches based on azimuth differences between successive points. Each branch corresponds to a specific azimuth range, allowing the system to handle different spatial relationships between points separately. This segmentation enables accurate predictions by considering local azimuth characteristics while maintaining an organized structure for efficient coding.
Solution Approach 2:
The patent applies local quality by creating prediction tree branches with specific azimuth thresholds for different regions of the point cloud. Instead of using a single uniform prediction model, the system creates localized prediction rules based on the azimuth differences observed in different parts of the data, improving prediction accuracy for each local region while maintaining overall coding efficiency.
2Measurement precision
If prediction tree branches are not terminated based on azimuth thresholds, then more points can be included in each branch, but prediction accuracy deteriorates when LIDAR does not rotate full 360 degrees
Solution Approach 1:
The prediction tree branches are dynamically terminated based on azimuth threshold comparisons. When the azimuth difference between successive points exceeds the threshold, the branch is terminated and a new branch is created. This dynamic termination strategy adapts the prediction tree structure to the actual spatial distribution of points, ensuring high prediction accuracy even when the LIDAR does not perform full 360-degree rotation, while still maximizing the number of points included in each branch.
Data Source
AI summary
An example device includes memory configured to store the point cloud data and one or more processors configured to determine a first point of the point cloud data to be a first node of a first prediction tree branch. The one or more processors are configured to determine that a first azimuth difference between the first point and a second point of the point cloud data does not meet a first azimuth threshold, and based on that determination, determine the second point to be a second node of the first prediction tree branch. The one or more processors are configured to determine that a second azimuth difference between a third point of the point cloud data and a fourth point of the point cloud data meets the first azimuth threshold and based on that determination, determine the fourth point to be a first node of a second prediction tree branch.