Target Detection in Images Using Adaptive Variance Calculation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing target detection methods in images struggle to accurately detect targets on multimodal backgrounds, leading to non-detection issues due to varying luminance values and high variance, which complicates environmental monitoring in fields like air and maritime surveillance.

Innovation Solution

A method that divides the image neighborhood into zones, calculates differences between zones, and adapts variance calculation formulas based on whether the background is monomodal or multimodal, allowing for accurate signal-to-noise ratio estimation and improved target detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional SNR calculation methods are used on multimodal backgrounds, then computational simplicity is maintained, but target detection accuracy deteriorates due to high variance and varying luminance values

Engineering Contradiction:
Improvetarget detection accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the background neighborhood into multiple zones (first background zone and second background zone) and calculates separate statistical parameters for each zone. This segmentation allows the method to handle multimodal backgrounds by treating different texture regions independently, thereby improving detection accuracy without requiring overly complex global analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different variance calculation approaches based on local background characteristics. By determining whether each local region is monomodal or multimodal and applying appropriate formulas, the method adapts to local conditions, improving accuracy while keeping computational requirements manageable through localized rather than global processing.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If the background is divided into multiple zones with separate variance calculations, then target detection accuracy on multimodal backgrounds is improved, but computational cost increases

Engineering Contradiction:
Improvesignal-to-noise ratio estimation accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent performs preliminary classification of background regions as either monomodal or multimodal before applying the appropriate variance calculation formula. This preliminary action allows the system to efficiently route processing through the correct computational path, avoiding unnecessary calculations and maintaining processing speed while improving accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes the variance calculation parameters based on the detected background type. By switching between different calculation formulas depending on whether the background is monomodal or multimodal, the system optimizes the balance between accuracy and computational efficiency for each specific scenario.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3959648B1Method for detecting targets
Publication Date: 2023.03.22 SAFRAN ELECTRONICS & DEFENSE (FR)
  • EP3959648B1 patent drawingFigure 1
  • EP3959648B1 patent drawingFigure 2A~2C
  • EP3959648B1 patent drawingFigure 3A~3B

AI summary

Method for detecting a target in an image made up of pixels, each pixel being represented by a value, comprising: obtaining (300) a position of a mask in said image, said mask covering at least one pixel of said image; obtaining (301) a predefined neighbourhood of pixels around said mask and dividing this neighbourhood into a plurality of zones; for each zone, calculating a mean (303) and a variance (304) of the values of the pixels of said zone; calculating (307), for each possible pair of zones, a value representative of a difference between said zones; determining whether a background of the mask is of the multimode or monomode type according to the values of the differences; calculating (308, 309, 310) a variance of the values of the pixels of said neighbourhood by using a formula for calculating the variance depending on the monomode or multimode type of said background; calculating a signal-to-noise ratio (311) for the position of the mask by using the variance thus obtained for the values of the pixels of said neighbourhood; and detecting (312, 313) a target when said signal-to-noise ratio is higher than a second predetermined threshold.