Video Object Segmentation via Multi-Layered Background Registration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video object segmentation methods face challenges in real-time processing due to the need for extensive calculation cycles and are prone to errors from non-uniform luminance and luminance shifts caused by light source changes, shading, and camera adjustments, making them unsuitable for efficient monitoring and recording in dynamic environments.
Innovation Solution
A multi-layered background registration method that stores static pixel data as background data in an image database, allowing for instant retrieval of a suitable background without additional calculation cycles, combined with the removal of average luminance from images and backgrounds to mitigate luminance-related issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional statistical calculation cycles are used to obtain background for object segmentation, then object segmentation can be performed, but processing time increases and real-time performance is lost
Solution Approach 1:
The patent pre-calculates and stores background information from multiple historical video frames before actual object segmentation is needed. By performing the statistical calculation cycle in advance and storing the results in a database, the system eliminates the need for time-consuming calculations during real-time segmentation, thus resolving the contradiction between segmentation accuracy and processing speed.
Solution Approach 2:
The patent accumulates multiple historical video frames and pre-processes them to create a robust background model that cushions against future variations. This pre-prepared background information acts as a buffer that allows rapid object segmentation without requiring real-time statistical calculations, thereby maintaining both accuracy and speed.
2Adaptability or versatility
If light source changes, shading variations, or camera adjustments occur, then image luminance shifts, but object segmentation accuracy deteriorates
Solution Approach 1:
The patent employs a dynamic background model that adapts to changing environmental conditions by incorporating multiple historical frames. Instead of using a static background, the system continuously updates and adjusts the background representation based on accumulated data, allowing it to dynamically respond to light source changes, shading variations, and camera adjustments while maintaining segmentation accuracy.
Solution Approach 2:
The patent changes the parameters used to represent the background by utilizing multiple historical video frames with different luminance characteristics. By storing and comparing across multiple frames, the system can identify consistent structural elements while filtering out transient luminance variations, thus maintaining segmentation accuracy despite environmental parameter changes.
3Adaptability or versatility
If multiple groups of cameras are periodically switched, then monitoring coverage is improved, but background retrieval timing is lost
Solution Approach 1:
The patent pre-accumulates and processes video frames from multiple camera groups in advance, building a comprehensive background model before switching occurs. By performing background extraction and storage proactively during periods when cameras are active, the system ensures that background information is ready for immediate retrieval when cameras switch, eliminating delays and maintaining both coverage and timing.
Solution Approach 2:
The patent creates copies of background information from multiple camera sources and stores them in a database. When cameras are periodically switched, the system can immediately retrieve the pre-copied background data corresponding to the active camera without waiting for new frames to be captured and processed, thus maintaining both multi-camera coverage and real-time performance.
Data Source
AI summary
A method and system of video object segmentation are disclosed herein. A pixel data of an image is received, wherein the pixel data has a corresponding location. A difference value between a pixel value of the pixel data and a pixel value in the corresponding location of a previous image is obtained. Utilizing the difference value and the pixel data, a multi-layered background registration is performed to get a background. Using the background, a background region of the image is removed. Furthermore, a process of removing an average value of luminance from the image and from the background is carried out to prevent object segmentation failure caused by the non-uniform luminance problem, which is produced by variation and flickering of illumination.