Point Cloud Geometry Encoding With Sensing Coverage for Missing Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current point cloud compression technologies face challenges in efficiently encoding and decoding point cloud geometry data, particularly in automotive applications where low latency and high compression efficiency are crucial, and existing solutions fail to accurately indicate missing data points, leading to potential safety issues in autonomous driving.
Innovation Solution
The method involves encoding and decoding point cloud geometry data using a bitstream that includes sensing coverage data, which represents the range of order indexes associated with sensed data, allowing for the indication of missing data points and improving decision-making in autonomous systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If point cloud compression is applied to reduce data transmission size, then bandwidth consumption is reduced, but accuracy of representing the original 3D scene deteriorates
Solution Approach 1:
The point cloud data is segmented into multiple levels of detail, with coarse geometry data providing the overall structure and fine geometry data providing detailed features. This hierarchical segmentation allows the system to transmit only the necessary level of detail based on application requirements, reducing bandwidth consumption while maintaining adequate accuracy for the specific use case
Solution Approach 2:
Different regions of the point cloud are encoded with different quality levels based on their importance. Critical regions (such as areas with moving objects or features of interest in autonomous driving) are encoded with higher precision, while less important regions use lower precision encoding. This local quality differentiation maintains accuracy where needed while reducing overall bandwidth consumption
2Loss of energy
If compression algorithms are made more complex to improve compression efficiency, then compression ratio is improved, but processing time increases
Solution Approach 1:
The encoding process performs preliminary organization of point cloud data into structured formats (such as octrees or hierarchical representations) before compression. This preliminary structuring enables more efficient compression algorithms to work on already-organized data, improving compression ratios without proportionally increasing processing time
Solution Approach 2:
The system applies compression algorithms selectively to different parts of the point cloud data based on priority. High-priority data (such as geometry information critical for safety decisions) receives more thorough compression processing, while lower-priority data uses faster, less intensive compression methods. This partial application of compression intensity optimizes the balance between compression efficiency and processing time
3Reliability
If all sensed data points are transmitted to ensure complete information, then data completeness is improved, but transmission bandwidth consumption increases
Solution Approach 1:
The system extracts and transmits only the essential features and critical data points from the complete sensed point cloud, rather than transmitting all raw data points. By identifying and extracting key geometric features, occupancy information, and salient objects, the system maintains data completeness for decision-making while significantly reducing transmission bandwidth consumption
Solution Approach 2:
The system transmits a partial representation of the point cloud data that is sufficient for the intended application (such as autonomous driving decisions) without transmitting the complete high-fidelity dataset. This partial transmission approach provides adequate reliability for safety-critical decisions while reducing bandwidth consumption by omitting redundant or less critical data points
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
There is provided methods and apparatus of encoding/decoding point cloud geometry data sensed by at least one sensor associated with a sensor index, point cloud geometry data being represented by ordered coarse points occupying some discrete positions of a set of discrete positions of a two-dimensional space, each coarse point being located within the two-dimensional space by an order index defined from a sensor index associated with a sensor that may sense a point of the point cloud represented by said coarse point and a sample index associated with a sensing time instant at which the point of the point cloud may be sensed. The encoding method comprises: - obtaining (110) a sensing coverage data (SCD) representative of at least one range of order indexes associated with sensed data; - encoding (120) said sensing coverage data (SCD) into the bitstream; and - encoding (130) point cloud geometry data into the bitstream.