Deep Learning Video Region Evaluation for Reliable Quality Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video quality monitoring systems struggle to accurately detect and validate quality loss in video streams due to impairments in communication channels and increased network traffic, leading to potential false alarms and unnecessary actions.
Innovation Solution
A system utilizing multiple deep learning models to evaluate regions within a video scene, employing different architectures and data sets to output quality scores and artifact types, with cross-validation to enhance detection confidence, and perform automated actions to mitigate quality loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple deep learning models with different architectures are used to evaluate video regions, then detection accuracy and confidence are improved, but device complexity increases
Solution Approach 1:
The video scene is divided into multiple regions, and different deep learning models are assigned to evaluate specific regions. This segmentation allows the system to achieve high detection accuracy through specialized models while managing complexity by distributing evaluation tasks across multiple smaller models rather than using one complex model for the entire scene.
Solution Approach 2:
Multiple deep learning models with different architectures are employed to perform the same quality loss detection function but with different evaluation approaches. This multi-functionality enables cross-validation of results, improving detection confidence while allowing the system to leverage the strengths of various model architectures for different types of artifacts.
2Reliability
If multiple deep learning models are deployed for cross-validation, then false alarms are reduced, but processing time increases
Solution Approach 1:
The system performs quality evaluation on multiple regions using multiple models, but not necessarily all models on all regions. The cross-validation is applied selectively to regions where quality loss is detected, allowing the system to reduce false alarms through validation while avoiding the excessive processing time that would result from evaluating every region with every model.
3Measurement precision
If deep learning models evaluate multiple regions, then quality loss detection confidence is improved, but computational resources are consumed
Solution Approach 1:
By dividing the video scene into multiple regions and assigning different deep learning models to evaluate specific regions, the system achieves high quality score accuracy through specialized regional evaluation. This segmentation reduces computational energy consumption compared to applying a single complex model to the entire scene, as each model processes only its assigned region.
Solution Approach 2:
Different deep learning models are deployed to evaluate different regions based on their specific characteristics and the types of artifacts expected in those regions. This local quality approach ensures that each region is evaluated by the most appropriate model, improving overall detection accuracy while optimizing computational resource usage by matching model capabilities to regional needs.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A system (200) for monitoring video quality and performing actions in response to validated quality issues. Detected video quality loss is validated by deep learning models operating on various regions of a video scene. Patterns of related regions are provided to the deep learning models and their quality scores and artifact types evaluated. Combining the quality scores of interrelated regions allows for increased confidence that the quality loss is significant, and actions should be taken based on the cross-validated results.