Signed Foreground Extraction Fusion for Video Abstract Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Surveillance video summarization is hindered by the poor blending of foreground objects into backgrounds due to light changes, resulting in low-quality video summaries and inefficient background updating processes that consume significant time and resources.

Innovation Solution

The method employs signed foreground extraction and fusion (S-FEF) combined with graphics processing unit (GPU) acceleration for background updating, using a single Gaussian model to separate static and dynamic frames and merge foreground data with updated backgrounds, preventing border shadows and enhancing processing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If conventional foreground extraction and background integration is used, then video summary generation is achieved, but border shadow artifacts appear due to poor blending of foreground objects into background

Engineering Contradiction:
Improvevideo summary qualityVSAvoidborder shadow artifacts
Core Design Contradiction:
Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The patent applies parameter changes by converting pixel data from unsigned 8-bit integer format to signed integer format. This parameter change in data representation enables the extraction of signed foreground data that preserves subtle intensity variations and gradient information, allowing for smoother blending of foreground objects into the background and eliminating border shadow artifacts through enhanced pixel-level integration.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If background updating algorithms are applied to adapt to light changes, then background quality improves, but processing time increases significantly due to large video data volume and algorithm complexity

Engineering Contradiction:
Improvebackground adaptation qualityVSAvoidbackground updating time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent substitutes the conventional CPU-based sequential processing mechanism with GPU-based parallel processing architecture. This mechanical substitution leverages the GPU's massively parallel compute units to perform background updating operations simultaneously across multiple pixel elements, dramatically reducing processing time while maintaining the quality of background adaptation to light changes.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transitions from single-threaded sequential processing to multi-threaded parallel processing by utilizing the GPU's architecture. This dimensional change in processing approach allows multiple background updating calculations to occur concurrently, transforming the time complexity from linear to near-constant relative to video data volume, thereby resolving the time-quality trade-off.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Quantity of substance

If surveillance video data volume is large, then comprehensive monitoring coverage is achieved, but storage requirements and processing burden increase significantly

Engineering Contradiction:
Improvevideo data volumeVSAvoidvideo summarization efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent extracts only the essential foreground information from complete video frames by identifying and isolating moving objects through background subtraction. This extraction process separates relevant motive objects from the full video data, creating a condensed representation that maintains monitoring coverage while dramatically reducing storage requirements and processing burden through selective data retention.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3340075B1Video abstract using signed foreground extraction and fusion
Publication Date: 2020.07.08 SAP SE
  • EP3340075B1 patent drawingFigure 1
  • EP3340075B1 patent drawingFigure 2
  • EP3340075B1 patent drawingFigure 3

AI summary

A video abstract is produced from surveillance camera video by retrieving a frame of the video, updating a background frame based at least in part on the retrieved frame, and performing a video division process which includes extracting, if it is determined that the average gray scale value of the retrieved frame exceeds a defined threshold, foreground data from the retrieved frame after conversion of the retrieved frame to a signed data type. The video division process further includes retrieving a next frame of the video produced by the camera, updating the background frame, and repeating the video division process. The producing of the video abstract further includes merging the foreground data with the updated background data after conversion of the foreground data and the updated background data to the signed data type, and generating a video abstract with the merged foreground and background data.