You are here

Profile-to-Profile Face Recognition Matching Capability

Description:

TECHNOLOGY AREA(S): Electronics, 

OBJECTIVE: The objective of this project is to develop/implement machine learning algorithms to automate identification of persons-of interest using face recognition with frontal and off-angle images from publically available information. 

DESCRIPTION: While there has been much focus on the use of Artificial Intelligence (AI) or Machine Learning (ML) based techniques to automate identification of persons-of-interest using face recognition using frontal and off-angle face images, there is a need to develop/augment a face recognition algorithm to be capable of matching faces at extreme angles, such as faces captured as full 90-degree profile images. Many operational use cases exist whereby DoD is only able to collect a profile face image (for example, images extracted from Captured Enemy Material and from publicly available information) and has the need to identify the unknown subject. The Project Manager (PM) is interested in algorithms that incorporate state-of-art AI and ML processes. Approaches should address use of neural network design and training processes; use of performance and model monitoring tools; and data analytics for validation and visualization. Preference is for solutions to be agnostic to future systems to allow rapid capability increase for fielded systems – in particular the US Army’s Video Identification, Collection and Exploitation (VICE) System. 

PHASE I: The objective is to develop overall system design that includes specification of AI/ML based techniques employed, specification of architecture required to support concept of operation, sensor specifications required to achieve match by various distances, recognition techniques employed by algorithm, and protocol for employment with current identity operations and intel platforms. 

PHASE II: Develop and demonstrate a prototype system in a realistic environment. Conduct testing to prove feasibility over extended operating conditions. 

PHASE III: Upon completion of research, software developed could be integrated into the Near Real Time Identity Operations (NRTIO) or Next Generation Biometric Collection Capability (NXGBCC) systems architecture. Providing the match/no-match response through a cloud-based architecture supports the cueing of multiple sensors on and off the battlefield. Technology could also be used to support smart cities concept in support of local governments via friction payment authentication or law enforcement applications. 

REFERENCES: 

1: Alyea, L.A., Hoglund, D.E., Eds. Human Detection and Positive Identification: Methods and Technologies, SPIE. 1996.

KEYWORDS: Machine Learning, Artificial Intelligence, Facial Recognition 

US Flag An Official Website of the United States Government