Image Processing Method for Removing Light Zones
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods, such as Gaussian Mixture Model (GMM), struggle to accurately detect moving objects at night due to areas of strong luminosity being misinterpreted as moving objects, leading to incorrect interpretations and difficulties in tracking vehicles or detecting objects illuminated by headlights.
Innovation Solution
An image processing method that filters light pollution by subtracting the background, determining a highlighting matrix to identify brighter pixels, limiting and normalizing pixel values, segmenting the matrix, and applying a mask to reduce light pollution areas, thereby enhancing the detection of moving objects or people.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Gaussian Mixture Model is applied to detect moving objects, then object detection capability is improved, but light zones are misinterpreted as moving objects leading to detection errors
Solution Approach 1:
The patent segments the image processing into distinct stages: background subtraction to obtain foreground, highlight matrix determination to identify bright pixels, compensation matrix creation to bound pixel values, and mask application to filter light zones. This segmentation allows separate handling of object detection and light pollution removal, resolving the contradiction by preventing light zones from being misinterpreted as moving objects while maintaining detection accuracy for actual objects.
Solution Approach 2:
The patent introduces intermediate data structures (highlight matrix, compensation matrix, mask) that act as mediators between the raw image data and the final detection result. These intermediaries process and filter the information sequentially, allowing the system to distinguish between actual moving objects and light zones before final detection, thereby improving both accuracy and reliability.
2Productivity
If background subtraction is performed to detect moving objects, then moving object detection is improved, but areas illuminated by headlights are incorrectly identified as objects
Solution Approach 1:
The patent performs preliminary actions before final object identification: first subtracting background to obtain foreground, then determining a highlight matrix to identify pixels with abnormal brightness, creating a compensation matrix to bound pixel values, and applying a mask to filter out light zones. These preliminary steps prepare the data by removing light pollution artifacts before the final detection stage, ensuring both speed and accuracy.
Solution Approach 2:
The patent extracts and removes the harmful component (light zones) from the image data through a series of operations: determining highlight matrices to identify bright pixels, creating compensation matrices to bound values, and applying masks to filter out extracted light zone regions. This extraction process separates actual moving objects from light pollution, improving identification accuracy while maintaining detection speed.
3Reliability
If highlight matrix determination is performed pixel by pixel to identify bright pixels, then light zone identification is improved, but processing complexity increases
Solution Approach 1:
The patent segments the complex pixel-by-pixel processing into modular stages: background subtraction, highlight matrix determination, compensation matrix creation, and mask application. Each stage handles a specific aspect of the problem, making the overall complex process more manageable and implementable while maintaining high reliability in light zone detection through systematic processing.
Data Source
AI summary
The subject of the present invention is an image-processing method for filtering light pollution appearing in a video stream of images acquired by a video camera. The method comprises, for a current image of said video stream of images, the steps of subtracting (E2) the background featuring in said current image in order to obtain the foreground of the current image, determining (E31) an over-brightness matrix, determining (E32) a compensation matrix by limiting the values of the pixels of the determined over-brightness matrix, segmenting (E33) the determined over-brightness matrix, determining (E34) a mask on the basis of the segmented over-brightness matrix, applying (E35) said mask to the determined compensation matrix in order to obtain a filtering matrix, and filtering (E37) the foreground of the current image by applying said filtering matrix in order to decrease the zones of light pollution in the images of the stream of images.