Dynamic Threshold Target Detection in Vehicle Imagery

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current target detection methods in images from optical sensors, especially for small targets like ballistic missiles, face challenges due to varying signal-to-noise ratios and registration errors, leading to inconsistent detection performance and high false alarm probabilities.

Innovation Solution

A method that estimates the background and statistical characteristics of the estimation error for each pixel, allowing for dynamic threshold determination based on local conditions to improve detection performance and reduce false alarms, using a Kalman filter and noise modeling to account for photonic and electronic noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a high-end optical sensor is used to reduce the dimensions of image portions, then the resolution is improved, but the complexity of the observation system and its cost increase

Engineering Contradiction:
ImproveresolutionVSAvoidcomplexity of the observation system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent uses image processing to create a processed version of the original image that enhances target visibility. Instead of relying solely on expensive high-resolution sensors, the system processes the captured images through algorithms that simulate the effect of higher resolution by enhancing edges and contrasting features, thereby achieving improved measurement precision without increasing device complexity

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms the image data by applying various processing parameters such as contrast enhancement, edge detection, and noise filtering. These parameter changes allow the system to extract target information more effectively from lower-resolution images, achieving the resolution improvement goal through software-based parameter optimization rather than hardware upgrades

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If a fixed threshold is used for target detection, then the detection process is simple, but the probability of false alarm varies with signal-to-noise ratio

Engineering Contradiction:
Improvesimplicity of detection processVSAvoidprobability of false alarm
Core Design Contradiction:
Ease of operationVSReliability

Solution Approach 1:

The patent implements dynamic threshold adjustment where the detection threshold is no longer fixed but adapts to local image conditions. The system calculates thresholds based on the statistical properties of the background and noise in different regions of the image, allowing the threshold to vary spatially and temporally. This dynamic approach maintains detection simplicity while significantly reducing false alarms by adapting to changing signal-to-noise ratios

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback mechanisms where detection results and image characteristics are fed back into the threshold determination process. By continuously monitoring the distribution of pixel values and detection outcomes, the system adjusts thresholds in real-time to maintain optimal detection performance, creating a closed-loop system that balances simplicity with reliability

Inventive Principle:
Principle #23Feedback

3Measurement precision

If background subtraction is performed to detect targets, then target visibility is improved, but detection performance varies with observed scene

Engineering Contradiction:
Improvetarget visibilityVSAvoiddetection performance consistency
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent applies local quality enhancement by performing background subtraction and thresholding operations that are adapted to local image characteristics. Instead of using global parameters, the system analyzes and processes each region according to its specific background properties, noise levels, and signal characteristics. This local adaptation ensures consistent detection performance across diverse observed scenes including different terrains, atmospheric conditions, and target types

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system performs preliminary background modeling and characterization before actual target detection. By pre-processing the images to establish baseline background statistics and characteristics, the system prepares the data in advance to facilitate more accurate and consistent target detection across varying scenes. This preliminary action reduces the impact of scene variability on detection performance

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP2360621B1Procedure for detecting a target in an image taken from a vehicle flying over a celestial body
Publication Date: 2013.10.16 ASTRIUM SAS
  • EP2360621B1 patent drawingFigure 1~3
  • EP2360621B1 patent drawingFigure 4~5
  • EP2360621B1 patent drawing

AI summary

The present invention relates to a method for detecting a target in at least one image consisting of a plurality of pixels, referred to as the "detection image," wherein the detection is performed by evaluating (30), in each pixel to be evaluated, a detection test against at least one threshold. The value of the at least one threshold depends on the pixel to be evaluated, and the method comprises, for each pixel to be evaluated, the steps of: - (10) estimating a background of the detection image at that pixel and estimating statistical characteristics of the error in estimating the background of the image at that pixel, - (20) determining the value of the at least one threshold at that pixel as a function of the estimated background of the image at that pixel and as a function of the estimated statistical characteristics of the error in estimating the background of the image at that pixel.