Contextual Video Resolution Modulation for Bandwidth Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
When network bandwidth drops during video streaming, existing technologies reduce video resolution, which can lead to a degraded user experience as important contextual objects become blurry, making it difficult for users to understand the content.
Innovation Solution
A system that applies machine learning to identify contextual and non-contextual objects in videos, generating separate streams to maintain the resolution of contextual objects even when the overall video resolution is reduced, ensuring that critical information remains clear.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If video resolution is reduced when bandwidth drops, then bandwidth consumption is decreased, but user experience is degraded as important contextual objects become blurry
Solution Approach 1:
The patent segments the video content by identifying and separating contextual objects from the rest of the video stream. Machine learning models detect and track contextual objects, creating distinct data structures for these objects versus the general video feed. This segmentation allows differential resolution handling where contextual objects can maintain higher resolution while the rest of the video is rendered at lower resolution, thus preserving important visual information while reducing overall bandwidth consumption.
Solution Approach 2:
The patent applies local quality by assigning different resolution qualities to different regions of the video stream. Specifically, contextual objects identified through machine learning are rendered at their original high resolution, while non-contextual areas are rendered at reduced resolution. This creates a spatially varying quality distribution that optimizes bandwidth usage while maintaining visual fidelity where it matters most for user understanding and engagement.
2Reliability
If machine learning is applied to identify contextual objects, then resolution of important objects is maintained, but system complexity increases
Solution Approach 1:
The patent introduces machine learning models as intermediary components that sit between the video stream and the rendering pipeline. These models act as mediators that analyze video frames, identify contextual objects, and generate metadata about object locations and significance. This intermediary layer enables intelligent decision-making about which regions require high resolution, automating what would otherwise require complex manual analysis and reducing overall system complexity despite the added processing step.
Data Source
AI summary
A method includes communicating a first stream of a video comprising first and second objects to a device. The first stream has a first resolution. The method also includes communicating a second stream to the device. The second stream indicates that the first object is contextual and that the second object is non-contextual. The method further includes, after a decrease in bandwidth, communicating a third stream of the video to the device. The third stream has a second resolution that is lower than the first resolution. When the video is presented for display using the third stream and based on the second stream indicating that the first object is contextual and that the second object is non-contextual, the first object is presented in the first resolution and the second object is presented in the second resolution.