Video Background Regeneration Using Block Correlation Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for regenerating the background of digital images from video streams captured by a fixed camera struggle when the scene includes partially stationary objects, as these methods often incorrectly integrate the stationary part of the object with the background, leading to processing defects and requiring an interference-free background image which is not always possible.
Innovation Solution
A method that cuts video frames and background images into blocks, calculates space and time correlations between these blocks, and updates the background image by selecting blocks with minimum correlation, using a weighted sum of absolute differences to ensure accurate background separation from foreground objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If time correlation methods are used to regenerate the background, then the processing can be performed with simple algorithms, but the stationary parts of foreground objects are incorrectly integrated into the background, polluting the regenerated background image
Solution Approach 1:
The image is divided into multiple blocks, and the background regeneration is performed block by block. For each block, the algorithm determines whether it belongs to the background or foreground by comparing with neighboring blocks, allowing precise handling of partially stationary objects without affecting the entire image.
Solution Approach 2:
Different processing strategies are applied to different blocks based on their local characteristics. Blocks with high correlation to neighbors are treated as background, while those with significant differences are treated as foreground, enabling localized adaptive processing that preserves accuracy while maintaining computational efficiency.
2Measurement precision
If an interference-free background image is required for accurate processing, then the background regeneration accuracy improves, but the requirement cannot be met in real-world scenarios where such images are not always available
Solution Approach 1:
The algorithm performs preliminary blocking and correlation analysis to identify potential background regions before final background regeneration. This preliminary action allows the system to prepare for accurate background extraction even when the initial background image contains interference from moving objects.
Solution Approach 2:
The algorithm uses the video stream itself to identify and regenerate the background without requiring external interference-free background images. By comparing temporal and spatial correlations within the video data, the system autonomously separates background from foreground, making it adaptable to real-world scenarios.
3Productivity
If the background image is continuously updated in real-time, then the processing latency is reduced and memory resources are optimized, but the accuracy of background separation may be compromised
Solution Approach 1:
By processing the image in blocks rather than the entire image at once, the algorithm achieves real-time updates with reduced computational burden. Each block can be independently and quickly evaluated against its neighbors, enabling fast background regeneration while maintaining accuracy through localized comparison.
Data Source
AI summary
The invention relates to a method for regenerating the background of digital images of a video stream comprising steps consisting in: —setting an initial background image, —cutting the unit images of the video stream into blocks b (i, j, t) and of the background image into corresponding blocks Bo (i, j, t). The method is essentially characterized in that it further includes steps consisting in: —selecting one block Bo of the background image and/or b of the frame image, and •calculating the space correlation thereof, with: •at least one block Bo of the background image at a time (t), and/or at another time (t−a), and/or •at least one block b of the frame image at a time (t), and/or at another time (t−a), and/or—updating the background image according to the calculation of the space correlation.