RGB and near-infrared aerial imagery dataset of a forested area with RTK-referenced UAV trajectories for visual SLAM and multispectral navigation development and benchmarking

Visual simultaneous localization and mapping (SLAM) is increasingly utilized for autonomous unmanned aerial vehicle operations in GNSS-degraded environments. However, there are limited public datasets that provide multispectral aerial imagery alongside the centimeter-level trajectory ground truth which is necessary for such algorithms, particularly in forested areas. This paper presents a dataset that addresses this deficiency. The study validates both the acquisition methodology and the benchmark quality of the resulting data. A custom RTK-referenced UAV was employed to capture time-synchronized RGB and near-infrared imagery over a 2.51 ha mixed coniferous-deciduous forest polygon in the Vilnius district of Lithuania, conducted across seven sessions in spring 2026 to reveal inter-day variations in illumination and weather conditions. Two intentionally distinct sets of cameras were utilized: an economical rolling-shutter camera pair and a global-shutter camera pair facilitate the evaluation of the same algorithm across varying shutter regimes and radiometric setups. Each session combines a photogrammetric grid mission with linear and diagonal localization transects at 80 meters above ground level. The release includes 629 raw and rectified RGB-NIR image pairs, intrinsic and stereo calibration, comprehensive per-frame telemetry, per-session METAR records, and mission waypoints, facilitating reproducible benchmarking of visual SLAM, visual odometry, and multispectral navigation in forested environments.