Abstract
We address the problem of whole-body person recognition in unconstrainedenvironments. This problem arises in surveillance scenarios such as those inthe IARPA Biometric Recognition and Identification at Altitude and Range(BRIAR) program, where biometric data is captured at long standoff distances,elevated viewing angles, and under adverse atmospheric conditions (e.g.,turbulence and high wind velocity). To this end, we propose FarSight, a unifiedend-to-end system for person recognition that integrates complementarybiometric cues across face, gait, and body shape modalities. FarSightincorporates novel algorithms across four core modules: multi-subject detectionand tracking, recognition-aware video restoration, modality-specific biometricfeature encoding, and quality-guided multi-modal fusion. These components aredesigned to work cohesively under degraded image conditions, large pose andscale variations, and cross-domain gaps. Extensive experiments on the BRIARdataset, one of the most comprehensive benchmarks for long-range, multi-modalbiometric recognition, demonstrate the effectiveness of FarSight. Compared toour preliminary system, this system achieves a 34.1% absolute gain in 1:1verification accuracy ([email protected]% FAR), a 17.8% increase in closed-setidentification (Rank-20), and a 34.3% reduction in open-set identificationerrors (FNIR@1% FPIR). Furthermore, FarSight was evaluated in the 2025 NIST RTEFace in Video Evaluation (FIVE), which conducts standardized face recognitiontesting on the BRIAR dataset. These results establish FarSight as astate-of-the-art solution for operational biometric recognition in challengingreal-world conditions.