STO-Activities: (no title)
Activity title:
Algorithms as Observers: The Future of Imaging System Design, Modeling, and Testing
Activity Reference:
SET-330
Panel:
SET
Security Classification:
NATO UNCLASSIFIED
Status:
Active
Activity type:
RTG
Start date:
2023-09-26T00:00:00Z
Actual End date:
2026-09-26T00:00:00Z
Keywords:
Aided Target Recognition, Algorithms, Computer Vision, EOIR, Imaging Systems, Machine Learning, SET, System Performance Modelling, System Testing
Background:
The existing design, modelling and testing methods for military imaging sensors focus on optimizing the performance for human observers. However, as the military modernization process increasingly incorporates robotic and automated systems, it is essential to adapt sensor design and analysis methods to prioritize algorithms as the primary consumers of imaging system data. This is a revolutionary transition that requires explicit research and development to assure that best practices and standards are used across NATO and allied nations. This proposal is not about algorithm development – but rather using algorithm performance and derived tools and metrics to perform design trades for future imaging systems specifically designed for machine speeds and algorithms as the first line consumer.
Objectives:
The objective of the RTG is to establish new practices and methods for imaging system design that prioritize algorithms as the primary consumer of sensor data. The central research will concentrate on developing predictive models for AiTR performance that can be used to perform design trades and develop requirements for target acquisition systems. However, a number of supporting topics are included in this activity, such as data collection and generation techniques, training data preparation, data storage and sharing, data annotation, target ontologies, algorithm scoring, and algorithm testing methods. Other related topics may also be considered as needed.
Initial research will focus on open, trained AiTR algorithms as baseline examples, measuring performance against various image degradation modalities, such as blur and noise. Using these examples, member nations will investigate various metrics and models that may predict performance against image degradations. The RTG will focus, at least initially, on EOIR imaging systems, ground-to-ground and air-to-ground targeting systems, and vehicles and dismount targets as test cases for metric and model development.
Using proposed models, there will be additional efforts with model verification, comparison to human performance, and testing of metrics and models with other target types and AiTR systems.
This RTG is not focused on algorithm development, so any data used in training baseline performance examples will use open data sets, and perhaps synthetic data sets, that allows the development of methods and standards without increasing the security level of the activity.
Topics:
Target acquisition performance, modeling, and testing of imaging systems that use algorithms for object detection, recognition, and tracking. All imaging wavelength bands are eligible with an emphasis on wavebands used by currently fielded targeting systems.
Discovery of metrics, signatures, feature vectors, and algorithmic methods useful for designing, modeling, measuring, and testing current and future algorithm-based imaging systems.
Developing software architectures, synthetic and virtual environments, automated design methods in support of current and future imaging systems that optimize algorithms and sensor as a combined system.
Development of new standards for modeling, measuring, and testing algorithm-based imaging systems.
Analysis of alternatives for future camera modalities that are best paired with algorithm based sensing designs.
Contact:
Open2Partners:
Title:
Created at 26/04/2023 17:00 by System Account
Last modified at 16/05/2024 21:00 by System Account
Go back to list
Home(NATO STO)