Live Point Cloud Compression via Egomotion Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for compressing live point cloud data, such as those used in autonomous driving applications, often rely on lossy video compression techniques that introduce significant errors in spatial range information, making them unsuitable for time-critical mapping applications.
Innovation Solution
The proposed solution involves a system that uses egomotion propagation to reduce bandwidth usage by comparing current range data with accumulated prior data, excluding similar data points and transmitting only necessary updates, thereby maintaining high accuracy and reducing data transmission requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If lossy video compression techniques are used to compress point cloud data, then bandwidth usage is reduced, but spatial range information accuracy deteriorates significantly
Solution Approach 1:
The system performs preliminary actions by accumulating prior range data and propagating it through egomotion estimation before compression. This allows the receiver to reconstruct missing data using propagated prior information, enabling more aggressive compression while maintaining accuracy. The preliminary propagation of range data creates a foundation that reduces the need to transmit all original data.
Solution Approach 2:
The system creates copies of prior range data through propagation and uses these copied versions at the receiver end to reconstruct the original point cloud. Instead of transmitting all original data, the system transmits only the differences or updates, with the receiver generating copies of the propagated data to fill in missing information, thereby reducing bandwidth while preserving accuracy.
2Measurement precision
If all current range data is transmitted to maintain accuracy, then spatial precision is preserved, but bandwidth consumption increases
Solution Approach 1:
The system extracts only the essential changes from current range data by comparing with propagated prior data. Instead of transmitting all data, it extracts and transmits only the differences or updates that cannot be reconstructed from propagated information. This extraction approach maintains spatial precision for critical data while significantly reducing bandwidth consumption by excluding redundant information.
Solution Approach 2:
The system applies partial action by transmitting only a subset of the full range data - specifically, only the portions that differ significantly from propagated prior data. This partial transmission strategy maintains necessary spatial precision while reducing overall bandwidth consumption by leaving out data that can be adequately reconstructed through propagation.
3Measurement precision
If live point cloud data is processed in real-time with strict requirements, then mapping accuracy is maintained, but computational complexity increases
Solution Approach 1:
The system performs preliminary egomotion estimation and range data propagation before the actual compression and transmission process. This preliminary action prepares the data in advance, creating a reference framework that simplifies subsequent real-time processing. By doing the heavy lifting of propagation beforehand, the system reduces computational complexity during real-time operation while maintaining mapping accuracy.
Solution Approach 2:
The system enables self-service by allowing the receiver to automatically reconstruct missing data using propagated prior information without requiring complex real-time processing of all transmitted data. The receiver uses the propagated data to fill in gaps and reconstruct the point cloud, reducing the computational burden on both transmitter and receiver while maintaining accuracy.
Data Source
AI summary
Approaches, techniques, and mechanisms are disclosed for range data compression. According to one embodiment, a current range data frame for a current time point is generated. Each current range data cell in the current range data frame includes current ranges representing points in a point cloud in a 3D space for the current time point. Accumulated prior ranges in an accumulated prior range buffer are propagated from a prior time point to the current time point. Current ranges in the current range data frame are compared with the propagated accumulated prior ranges to determine range differences between the current ranges and the propagated accumulated prior ranges. A proper subset of current ranges in the set of current ranges is identified based on the range difference. The proper subset of current ranges is included in a range data signal excluding other current ranges not in the proper subset.