Dynamic Region of Interest Construction for Video Surveillance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems face inefficiencies in image storage and processing due to the arbitrary shape of predefined regions of interest, which are not adapted to the actual scene configuration, leading to increased storage costs and processing loads.
Innovation Solution
A method and system for constructing regions of interest by analyzing images to identify optimal areas of any shape, based on image analysis and task-specific criteria, allowing for dynamic selection and updating of relevant image portions to guide camera settings and reduce unnecessary processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If regions of interest are defined by users based on one or a few images, then the region definition process is simple, but the region shape is arbitrary and not adapted to the actual scene configuration
Solution Approach 1:
The image is automatically segmented into multiple candidate regions based on scene analysis, target detection, and semantic understanding. Each candidate region is evaluated for its relevance to the surveillance task, allowing the system to identify optimal region shapes that adapt to the actual scene configuration rather than using arbitrary user-defined shapes.
Solution Approach 2:
The system performs self-service by automatically analyzing the scene and generating optimized region of interest definitions without requiring manual user input. The camera settings and region boundaries are automatically adjusted based on the analyzed scene characteristics, eliminating the need for users to manually define regions while achieving adaptive shape optimization.
2Quantity of substance
If all images are analyzed before storage to identify images for recording, then the amount of data to be stored is reduced, but the processing load and resource consumption increase
Solution Approach 1:
The system extracts only the essential information from images during analysis - specifically identifying candidate regions of interest and their characteristics. Instead of analyzing entire images for storage decisions, the system extracts region-level features and uses these condensed representations for subsequent processing, reducing both storage requirements and processing loads.
Solution Approach 2:
The system performs preliminary scene analysis and region identification before full image processing and storage decisions are made. By pre-identifying candidate regions and their relevance to surveillance tasks, the system prepares optimized region definitions that guide subsequent image analysis and storage operations, improving overall processing efficiency.
3Quantity of substance
If regions of interest are used to monitor motion and determine image storage, then storage costs are reduced, but the regions have predetermined arbitrary shapes that are not adapted to the actual configuration
Solution Approach 1:
The region of interest boundaries are made dynamic rather than static. The system continuously analyzes scene changes and automatically adjusts region shapes to match the actual configuration of relevant objects and areas. This dynamic adaptation allows regions to change shape based on scene content while maintaining the storage efficiency benefits of region-based monitoring.
Solution Approach 2:
The system changes the geometric parameters of regions based on scene analysis results. Instead of using fixed predetermined shapes, the region boundaries, areas, and positions are dynamically adjusted according to detected targets, scene semantics, and surveillance requirements, achieving both storage efficiency and geometric adaptability.
4Reliability
If VCA is applied to all images, then comprehensive analysis is achieved, but processing power is wasted on analyzing irrelevant parts of images
Solution Approach 1:
The system applies different processing qualities to different regions of the image. Video Content Analytics are concentrated on the identified candidate regions of interest, while irrelevant areas receive minimal or no processing. This local quality differentiation maintains reliable analysis of important areas while significantly reducing overall processing power consumption.
Solution Approach 2:
Instead of applying VCA uniformly to all images, the system applies partial action by limiting detailed analytics only to the relevant candidate regions identified through preliminary scene analysis. This selective application of processing power maintains analysis reliability for important areas while avoiding waste on irrelevant portions of images.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
At least one embodiment of a method for constructing a region of interest from images representing a same scene, the region of interest being used for setting the camera used to obtain the images and/or to process images obtained from the camera, the method comprising: obtaining (300) a plurality of images representing a same scene; detecting (310) predetermined targets in images of the plurality of images; in response to detecting predetermined targets, segmenting (315) an area corresponding to the images of the plurality of images into portions of image, each of the portions being associated with a relevance indication for the corresponding portion to be selected as part of a region of interest; selecting (325) at least one relevant portion among the portions, as a function of the relevance indications; upon detecting selection of a relevant portion, updating (330) the relevance indication associated with relevant portions different than the selected relevant portions, as a function of the selected relevant portions; and constructing (340) a region of interest based on the selected relevant portions.