Video Analytics Parameter Sharing Across Similar Camera Classifications
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video analytics systems are limited in handling atypical scenarios and require site-specific learning, which is local and applies only to individual cameras, lacking a method to share and update video analytics parameters across similar camera configurations.
Innovation Solution
A computing device communicates with multiple cameras and video analytics engines, classifying cameras by scene type and updating video analytics parameters across similar scenes, using feedback to refine and share parameters, enabling distributed learning and improved recognition capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If video analytics parameters are configured individually for each camera at installation, then each camera can be customized for its specific scene, but the system lacks the ability to share and propagate parameter improvements across similar cameras
Solution Approach 1:
The patent combines individual camera parameter learning with system-wide parameter sharing by maintaining a centralized parameter store that aggregates learning from multiple cameras. When a camera learns improved parameters for its scene classification, these parameters are merged into the centralized store and automatically propagated to other cameras with matching scene classifications, resolving the contradiction between individual customization and system-wide information sharing.
Solution Approach 2:
The patent creates universal parameter sets for each scene classification that can be applied across multiple cameras regardless of their individual locations. The system determines scene classifications for different cameras and applies the same refined parameters universally to all cameras sharing that classification, enabling one parameter set to serve multiple functions across the entire video analytics system.
2Ease of manufacture
If video analytics systems use out-of-the-box parameters for typical scenarios, then installation is simplified and faster, but the system cannot effectively handle atypical or site-specific scenarios
Solution Approach 1:
The patent applies preliminary action by pre-configuring cameras with out-of-the-box parameters appropriate for typical scenarios during installation, enabling immediate operational capability. Subsequently, the system performs preliminary learning during operational phases to refine these parameters for atypical or site-specific scenarios, and then propagates these refinements system-wide, thus maintaining both installation simplicity and adaptability to special cases.
Solution Approach 2:
The patent implements self-service by enabling the video analytics system to automatically learn and refine its own parameters through feedback mechanisms. The system monitors detection accuracy and false positives, automatically adjusts parameters for its scene classifications, and propagates improvements without requiring manual reconfiguration, thus maintaining ease of installation while achieving high adaptability.
3Reliability
If site-specific learning is implemented for individual cameras, then local performance improves over time, but the learning remains isolated and does not benefit other cameras with similar configurations
Solution Approach 1:
The patent implements feedback mechanisms that collect performance data from individual cameras regarding detection accuracy and false positives. This feedback is aggregated at the system level, allowing the system to identify parameter improvements based on collective performance data. The refined parameters are then propagated back to all cameras with matching scene classifications, transforming isolated local learning into system-wide productivity improvement while maintaining local reliability.
Data Source
AI summary
A device, method and system for installing video analytics parameters at a video analytics engine is provided. An example device determines that a classification of a first camera is one or more of similar to, and same as, a respective classification of at least one second camera. The example device retrieves, from a memory, video analytics parameters associated with the at least one second camera, the video analytics parameters stored at the memory in association with the respective classification. The example device causes installing of the video analytics parameters at a video analytics engine associated with the first camera.