Sensor-Indexed Point Cloud Coding for Low-Latency Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud codecs fail to combine simplicity, low latency, and high compression performance, particularly when encoding and decoding sparse geometry data sensed by sensors with flexible sensing paths, such as those used in autonomous vehicles, due to inefficient predictive coding and lack of adaptation to the order of sensing.
Innovation Solution
A method and apparatus for encoding and decoding point cloud geometry data using a coarse representation in a two-dimensional space, where each point is associated with a sensor index and a sample index, allowing for predictive encoding of a residual radius based on candidate radii from different sensors and time instants to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If traditional predictive coding is used for point cloud geometry data, then encoding simplicity is maintained, but compression performance deteriorates due to large dynamic range of residual radii
Solution Approach 1:
The patent applies preliminary action by constructing a list of candidate radii before performing predictive encoding. The encoder pre-processes the data to identify potential predictor radii from previously encoded points, then selects the best candidate from this pre-prepared list. This preliminary preparation enables more accurate prediction without significantly increasing encoding complexity, thereby reducing the dynamic range of residual radii and improving compression performance.
2Loss of substance
If complex predictive coding schemes are implemented to improve compression performance, then bitrate is reduced, but encoding and decoding latency increases
Solution Approach 1:
The patent implements partial action by considering only a limited number of candidate radii from the list rather than exhaustively searching all possible predictors. The encoder selects from a manageable subset of candidate radii that are most likely to provide good prediction, achieving sufficient compression performance without the computational overhead of exhaustive search. This partial approach maintains low encoding latency while still improving compression compared to traditional methods.
3Loss of substance
If traditional occupancy-based coding is used, then sparse geometry data is handled, but compression performance deteriorates due to lack of adaptation to sensing order
Solution Approach 1:
The patent applies dynamics by adapting the predictive coding scheme to the specific sensing path and order of the LiDAR scanner. Instead of using a fixed coding structure, the encoder dynamically selects candidate radii based on the actual sensing sequence and spatial relationships observed in the data. This dynamic adaptation to the sensing order improves compression performance for sparse geometry data by exploiting the specific patterns introduced by the sensor's scanning mechanism.
4Measurement precision
If more candidate radii are considered for prediction, then prediction accuracy improves, but encoding complexity increases
Solution Approach 1:
The patent implements partial action by considering only a limited number of candidate radii from the list rather than exhaustively searching all possible predictors. The encoder selects from a manageable subset of candidate radii that are most likely to provide good prediction, achieving sufficient compression performance without the computational overhead of exhaustive search. This partial approach maintains low encoding latency while still improving compression compared to traditional methods.
Data Source
AI summary
A method of encoding, into a bitstream, point cloud geometry data sensed by at least one sensor associated with a sensor index, includes: obtaining a list of candidate radiuses (L={r2}) for a first occupied coarse point (P1) having a first sensor index (λ1), a first sample index (s1) and being associated with a first radius (r1); selecting, for the first occupied coarse point (P1), a selected predicting radius (rpred) from the list of candidate radiuses (L={r2}); encoding, in the bitstream, a data (Ipred) representative of the selected predicting radius (rpred) in the list of candidate radiuses (L={r2}); and predictively encoding, in the bitstream, a residual radius (rres) between the first radius (r1) and the selected predicting radius (rpred).