Quantifying Operational Drivers of Multimodal Biometric Verification in Aerial Surveillance

Alina Peluso, David S. Bolme, Gavin Glenn, Jairus Hines, Nicholas Burchfield, Andrew Duncan, Youngrock Yoon, Joel Robert Gray Brogan, Deniz Aykac, Ryan Shivers, Gio Pascascio, Leanne Thompson, Scott dolvin, David Cornett

Track: Proceedings Track

Keywords: multimodal biometric verification, aerial surveillance, unmanned aerial systems (UAS), cross-domain verification, BRIAR)

TL;DR: This paper quantifies how operational factors such as image resolution, sensor type, and algorithm choice drive multimodal biometric verification performance in aerial (UAS) versus close-range surveillance video.

Abstract: Multimodal biometric verification is increasingly applied across operational contexts ranging from close-range security cameras and building-mounted surveillance to long-range ground sensors and unmanned aerial system (UAS) imagery. Variations in acquisition conditions—such as image resolution, viewing geometry, and motion artifacts—pose significant challenges for cross-domain algorithmic generalization. This study evaluates two independent multimodal biometric verification systems developed under the Intelligence Advanced Research Projects Activity (IARPA) Biometric Recognition and Identification at Altitude and Range (BRIAR) program, comparing performance on close-range and aerial datasets. Close-range video served as a baseline to quantify the decline in verification performance on aerial footage. The dataset included six UAS platforms, spanning small quadcopters at 10m altitude to medium-sized fixed-wing aircraft at 360m. Mixed-effects logistic regression identified image resolution (head and body pixel counts), head height, sensor characteristics, and algorithm selection as primary determinants of verification success, whereas demographic attributes and mission gait were not significant predictors. Activity type and collection site influenced performance in close-range data but had negligible impact on UAS imagery. These results clarify modality-specific strengths and limitations and highlight opportunities to enhance cross-domain biometric verification.

Submission Number: 6