Hybrid Projection Point Cloud Texture Coding for Occlusion Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional point cloud compression technologies face inefficiencies in encoding moving or deforming objects, leading to data loss and distortion due to occlusion issues in 2D projections, which affects the accuracy and bandwidth utilization of 3D geometrical representations.
Innovation Solution
A hybrid projection-based method that generates 2D projections from 3D geometrical representations and distinctly encodes occluded points, using 2D image encoding techniques for projections and 3D object encoding techniques for occluded points, allowing for efficient compression without quality degradation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a 2D projection approach is used for point cloud compression, then compression efficiency is improved, but information loss occurs due to occlusion of points
Solution Approach 1:
The patent segments the point cloud into visible points and occluded points based on 2D projection analysis. Visible points are encoded using efficient 2D projection methods, while occluded points are separately identified and encoded using 3D spatial relationships, allowing each segment to be handled with the most appropriate encoding strategy.
Solution Approach 2:
The patent applies different encoding qualities and methods to different regions of the point cloud. Visible regions use compressed 2D projection encoding, while occluded regions use more robust 3D-based encoding to preserve critical information that would otherwise be lost, ensuring local optimization of encoding quality where needed.
2Stability of the object's composition
If a conventional 3D encoding architecture is used, then spatial relationships are preserved, but compression efficiency deteriorates for moving or deforming objects
Solution Approach 1:
The patent dynamically adapts the encoding approach based on the content being encoded. By analyzing the 2D projection, the system dynamically identifies occluded points and switches between 2D projection-based encoding for visible points and 3D spatial-based encoding for occluded points, optimizing compression efficiency for moving and deforming objects while preserving necessary spatial relationships.
3Loss of information
If all points in the 3D point cloud are encoded, then complete information is preserved, but processing power and bandwidth consumption increase
Solution Approach 1:
The patent extracts and separately handles occluded points from the main point cloud data. By identifying points that would be occluded in 2D projections and extracting them for separate encoding, the system avoids transmitting redundant information while preserving critical occluded point data, thereby reducing overall processing power and bandwidth consumption.
Solution Approach 2:
The patent applies partial encoding action by selectively encoding only the necessary points. Instead of encoding all points with equal detail, it uses 2D projection encoding for visible points (which is more efficient) and applies additional 3D encoding only to occluded points where information preservation is critical, optimizing the balance between information completeness and resource consumption.
Data Source
AI summary
An apparatus receives a video which includes at least one three-dimensional (3D) object in a 3D physical space. A 3D geometrical representation of a point cloud is generated based on the video. The 3D geometrical representation of the point cloud includes a first set of points associated with geometrical information and texture information corresponding to the at least one 3D object. A plurality of two-dimensional (2D) projections are generated from the 3D geometrical representation of the point cloud. A second set of points that are occluded in the first set of points is detected, corresponding to the plurality of 2D projections. The plurality of 2D projections and the second set of points are distinctly encoded, and the remaining points, other than the detected second set of points, in the first set of points are discarded for efficient compression of the 3D geometrical representation of the point cloud.