Event Camera Stream Compression Using Field-Specific Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Event streams from event-based cameras require efficient compression methods that consume minimal resources, especially in low-power environments, while maintaining data compactness, as conventional tools like Zip and Gzip are resource-intensive.
Innovation Solution
A compression method utilizing run-length encoding for timestamps and y-coordinates, dictionary encoding for x-coordinates, and bit-packing for polarities, tailored to specific event stream structures, reducing resource consumption and achieving efficient data compression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional generic compression tools (Deflate, Zip, Gzip) are used on event streams, then compression results are satisfactory, but resource consumption is significant
Solution Approach 1:
The event stream data is segmented into distinct fields (timestamp, y-coordinate, x-coordinate, polarity) and each field is compressed using a specialized algorithm optimized for its specific data characteristics. This segmentation allows tailored compression strategies for each field type, achieving efficient compression with lower resource consumption compared to generic tools.
Solution Approach 2:
The invention changes the parameter representation format from standard 64-bit floating point to compact integer encodings. Timestamps are converted to 32-bit integers, coordinates to 16-bit integers, and polarity to 1-bit values. This parameter transformation significantly reduces data size and enables more efficient processing with lower resource usage.
2Quantity of substance
If event data is stored in raw format, then storage efficiency is reduced, but processing simplicity is maintained
Solution Approach 1:
The compression algorithm performs preliminary actions by pre-defining optimal encoding schemes for each event field type. Run-length encoding is pre-configured for timestamp and coordinate fields, dictionary encoding is pre-prepared for polarity sequences, and bit-packing formats are pre-established. This preliminary preparation enables efficient compression without requiring complex real-time decision-making during processing.
Solution Approach 2:
The invention uses copying techniques by creating compact representations of event data that replicate the essential information in a more efficient format. Instead of storing full 64-bit floating point values, it copies only the necessary precision into compact integer and bit fields, achieving space efficiency while maintaining data integrity.
Data Source
AI summary
A method produces a compressed sequence of events stored in a memory, originating from an event-based camera, the sequence comprising a first run-length code representing timestamps of the sequence of events; and subsequent compressed codes including one of: (i) a second run-length code representing first coordinates of the sequence of events, at least one dictionary code representing second coordinates of the sequence of events, and data representing polarities of the sequence of events; (ii) a second run-length code representing first coordinates of the sequence of events, a number of bit-packed fields, each representing a second coordinate and polarity pair of the sequence of events; or (iii) a number of bit-packed fields, each representing a first and second coordinate and polarity triplet of the sequence of events.