Region-Packed Video Compression for Machine Vision Data Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding technologies are inefficient for applications where only specific objects or regions of interest need to be compressed for machine consumption, as they do not effectively discard irrelevant data and optimize compression based on these regions.
Innovation Solution
A video encoding and decoding system that detects regions of interest, tightly packs them into contiguous areas, and omits non-relevant data, using bounding boxes to preserve object location and shape during compression and decompression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional video coding compresses entire frames including all objects and background, then complete visual information is preserved, but data volume is large and compression efficiency is low for machine-only consumption
Solution Approach 1:
The patent extracts only the regions of interest (objects) from the video frame and discards the background and irrelevant areas. This is achieved by detecting object boundaries and creating masks that isolate only the necessary visual information for machine consumption, thereby reducing data volume while eliminating irrelevant data.
Solution Approach 2:
The video frame is segmented into multiple regions: regions of interest (objects to be tracked), background regions, and transition regions. Each region is processed differently - objects are preserved with their features, background is discarded, and transition regions are handled with special blending techniques. This segmentation enables selective compression based on importance.
2Productivity
If only regions of interest are compressed and background is discarded, then compression efficiency improves, but object location and shape precision may deteriorate
Solution Approach 1:
Different quality levels are applied to different regions of the frame. Regions of interest receive high-quality encoding with preserved spatial resolution and detailed features, while background regions are completely discarded. This local quality differentiation maintains precision for objects while maximizing compression efficiency.
Solution Approach 2:
The patent transforms the 2D spatial information into a packed representation where multiple objects are arranged in a new coordinate system. Objects are tightly packed to eliminate empty space, and their original positions are reconstructed using stored position metadata. This dimensional transformation achieves better compression while preserving location precision through metadata.
3Quantity of substance
If objects are tightly packed to eliminate empty space, then data volume reduces, but complexity of rearranging and reconstructing regions increases
Solution Approach 1:
The patent performs preliminary actions during encoding by detecting objects, determining their boundaries, and calculating their packed positions before actual compression. Position metadata and transformation parameters are pre-computed and stored, so the decoder only needs to apply simple transformations rather than complex rearrangements, reducing decoding complexity.
Solution Approach 2:
The patent introduces an intermediary packed representation format that serves as a bridge between the original frame and the final compressed output. Objects are first extracted, then packed into this intermediate format with stored position information, and finally reconstructed by applying the stored transformations. This intermediary format simplifies the overall process by separating extraction, packing, and reconstruction into manageable stages.
Data Source
AI summary
Systems and methods for video coding and decoding using region packing are provided. At an encoder, a region detection module receives a video frame for encoding, identifies regions of interest in the video frame, and generates a bounding box for each region of interest. A region extractor module obtains the pixels within the bounding box from the video frame. A region packing module receives the identified regions of interest and arranges the bounding boxes within a packed frame substantially reducing the data to be encoded outside the identified regions of interest. A video encoder receives the packed frame and generates an encoded bitstream therefrom. At the decoder, the encoded bitstream is decoded and parameters sufficient to place the regions within a reconstructed frame are extracted. A reconstructed frame is generated which substantially maintains the spatial relationship and size of regions of interest in the original video frame.