Skip to main content
Initial Planning Point | IPP Search smarter
← All articles Research summary

Human detection in UAV-captured imagery for search and rescue operations: a aystematic review of deep learning and image processing methods

PRISMA-guided review of 48 studies finds YOLO-family models dominate benchmark performance, but documented operational deployment remains limited.

Read the original source

Published on IPP

  • Author(s): Thomas Oliver de Aquino; Alexandre Carlos Rodrigues Ramos; Felix Mora-Camino; Oludolapo Akanni Olanrewaju; Alexandre Carlos Brandão Ramos

  • Source: CEAS Aeronautical Journal

  • Publisher(s): Springer Nature

  • Publication date: 29 October 2026

Why it matters in operational practice: High precision or recall in a dataset is not a field POD. Treat detections as prompts for trained human verification and log environmental conditions, altitude, imagery quality and false alerts.

Operational transferability: Relevant across UAV-enabled SAR, with local validation needed for terrain, sensor, tasking, airspace and imagery-review workflows.

Important methodological limitations: Heterogeneous datasets and metrics prevent reliable cross-model comparison; few studies evidence live operational use.

Training influence: Build exercises that require crews to verify alerts, record false positives/negatives and explain how negative aerial searches change tasking and assurance.

SAR training sections: UAV operations; visual search; POA/POD/cumulative probability; resource selection and tasking; Search Operations and progress monitoring.