Search papers, labs, and topics across Lattice.
The paper introduces ForestPersons, a new large-scale dataset designed for training and evaluating person detection models specifically for under-canopy forest environments. The dataset contains 96,482 images with 204,078 annotations, including bounding boxes, pose, and visibility labels, captured from ground-level and low-altitude perspectives relevant to Micro Aerial Vehicle (MAV) search and rescue operations. Baseline evaluations demonstrate that existing object detection models trained on general or SAR-oriented datasets perform poorly on ForestPersons, highlighting the dataset's unique challenges and its potential to advance person detection in real-world SAR scenarios.
Object detectors trained on standard datasets struggle to find people under forest canopies, motivating a new benchmark tailored to the challenging visual conditions faced by search and rescue drones.
Detecting missing persons in forest environments remains a challenge, as dense canopy cover often conceals individuals from detection in top-down or oblique aerial imagery typically captured by Unmanned Aerial Vehicles (UAVs). While UAVs are effective for covering large, inaccessible areas, their aerial perspectives often miss critical visual cues beneath the forest canopy. This limitation underscores the need for under-canopy perspectives better suited for detecting missing persons in such environments. To address this gap, we introduce ForestPersons, a novel large-scale dataset specifically designed for under-canopy person detection. ForestPersons contains 96,482 images and 204,078 annotations collected under diverse environmental and temporal conditions. Each annotation includes a bounding box, pose, and visibility label for occlusion-aware analysis. ForestPersons provides ground-level and low-altitude perspectives that closely reflect the visual conditions encountered by Micro Aerial Vehicles (MAVs) during forest Search and Rescue (SAR) missions. Our baseline evaluations reveal that standard object detection models, trained on prior large-scale object detection datasets or SAR-oriented datasets, show limited performance on ForestPersons. This indicates that prior benchmarks are not well aligned with the challenges of missing person detection under the forest canopy. We offer this benchmark to support advanced person detection capabilities in real-world SAR scenarios. The dataset is publicly available at https://huggingface.co/datasets/etri/ForestPersons.
