Two-Dimension Point Cloud Encoding Using Unfolded Grid Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for encoding and decoding two-dimension point clouds face challenges in efficiently optimizing the distribution of points over a frame, particularly in distinguishing between close points encoded in the same pixel and computing the inverse mapping operator, which is resource-intensive and often not invertible.
Innovation Solution
The method involves generating an unfolded grid using a dense mapping operator and applying a piecewise bilinear interpolator to optimize the distribution of points, minimizing unused pixels and maximizing information pixels, allowing for efficient encoding and decoding by parametrizing the un-mapping operator for efficient data storage and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a complex mapping operator is used to optimize point distribution over the frame, then the distribution of points is improved and unused pixels are minimized, but the data size required to encode the mapping operator increases and processing resources are consumed
Solution Approach 1:
The patent transforms the complex mapping operator into a parametrized function by defining it through control points and basis functions. This allows the operator to maintain its point distribution optimization capability while being represented by a compact set of parameters (control point coordinates and function coefficients) rather than a complex algorithm, thereby reducing encoding data size and processing requirements
Solution Approach 2:
The patent uses a regular grid as a simplified copy or approximation of the optimal point distribution. By mapping points to a regular grid structure and using interpolation to achieve smooth distribution, the system avoids needing to encode and process a fully complex mapping operator, thus reducing complexity while maintaining acceptable point distribution quality
2Measurement precision
If the inverse of the mapping operator is computed at the decoding side to retrieve original point cloud, then the original coordinates can be retrieved, but time and processing resources are consumed
Solution Approach 1:
The patent inverts the problem approach by designing the mapping operator to be inherently invertible through its parametrized structure. Instead of computing a complex inverse operator, the system uses the same parametrized function form at the decoder side with the transmitted parameters, allowing efficient reconstruction of original coordinates without time-consuming inverse computation
Solution Approach 2:
The patent prepares the mapping operator in advance as a parametrized function during encoding, and transmits these parameters to the decoder. This preliminary preparation allows the decoder to efficiently reconstruct points using the same parametric formulation without needing to compute complex inverses in real-time, thus reducing decoding time
3Device complexity
If some close points are encoded in the same pixel, then the encoding process is simplified, but it is no longer possible to distinguish these points at the decoding side
Solution Approach 1:
The patent resolves the point distinction problem by introducing additional dimensional information through the parametric mapping space. Even when multiple points map to the same pixel, their original coordinates can be recovered by using the parametric function to compute their positions in the extended parameter space, thus preserving point distinction information without complicating the pixel encoding process
Data Source
AI summary
The present disclosure relates to methods, devices or streams for encoding, transmitting and decoding two-dimension point clouds. When encoding point clouds as frames, a large number of pixels are not used. A dense mapping operator optimizes the use of pixels but requires a lot of data to be encoded in the stream and its inverse operator is difficult to compute. A simplified mapping operator is generated according to a dense mapping operator and is stored as matrices of two-dimension coordinates representative of an unfolded grid which requires low space in the stream. The inverse operator is easy to generate according to the unfolded grid.