UAVLoc-M3 UAV Visual Localization Dataset

The dataset was collected by Sony ILCE-7RM2 and Sony ILCE-6000 aerial survey cameras mounted on a drone platform at intervals of flight time in real scenes, covering drone images from three regions in China. The sample locations include Chongming Island in Shanghai, Anyang River Basin in Henan Province, and Cuijiaqiao Town in Anyang County. The drone flies in the air at a relative altitude range of 300 meters to 800 meters and samples images from different flight attitude perspectives. Its flight scenes pass through various terrains and landforms such as villages, towns, rivers, and farmland, covering both spring and summer seasons, as well as sunny and foggy weather conditions.We choose Bing Satellite Maps to define the latitude and longitude range of the satellite map based on the drone's flight coverage. In order to ensure a fixed ratio between satellite image resolution and drone image resolution captured at different heights, we used the LocaSpace Viewer software platform to obtain image data from Bing satellite maps at Zoom Level 17 and Zoom Level 18 levels. The ground resolution of satellite images at level 17 is approximately 0.965 meters per pixel; The ground resolution of satellite images at level 18 is approximately 0.508 meters per pixel.Because the visual axis direction of the camera can be affected by the drone's attitude and deviate directly downwards, directly using the drone's own GPS coordinates will introduce significant errors. We use an algorithm based on geometric models and digital elevation models (DEM) to accurately estimate the geographic coordinates of the intersection point between the drone camera lens axis and the ground. The UAVLoc-M3 dataset contains latitude and longitude coordinates (lon, lat), altitude (GPs_alt), and attitude angles (roll, pitch, yaw) recorded by unmanned aerial vehicle GNSS receivers. We also obtained accurate ground elevation (ground_alt) by querying the 30 meter resolution digital elevation model (DEM) data provided by the Space Shuttle Radar Topography Mission (SRTM). We calculate the attitude angle of the drone and correct the latitude and longitude positions of the actual terrain changes obtained from DEM at the shooting point, in order to obtain the center position of the camera field of view.For the files in the dataset, they mainly include large-scale satellite images, drone images, relevant information recorded during drone shooting, and digital elevation data. Among them, satellite images are in JPG format; The drone image is in JPG format and the resolution is uniformly processed to 320 × 320; The recorded information is in XLSX table format; The digital elevation data is in. hgt format. The dataset includes three folders named AnYangRiver, Chongmingdao, and CuiJiaQiao, representing datasets from three different regions: Anyang River Basin, Chongming Island, and Cuijiaqiao Town. Each folder contains folders DEM (digital elevation data), drone (drone images), satellite (satellite images), and record information tables. The satellite images of Chongming Island area are divided into 11 different parts of the images.For the record information table of unmanned aerial vehicles, each table contains 7 columns, and each row corresponds to the position and attitude information of the unmanned aerial vehicle at the time of image capture, which are lon (longitude/degree), lat (latitude/degree), GPS-Alt (altitude/meter), roll (roll angle/degree), pitch (pitch angle/degree), yaw (azimuth angle/degree). The number of data items in the table is the same as the number of drone images in the corresponding area, including 644 AnYangRiver.xlsx, 3695 Chongmingdao.xlsx, and 607 CuiJiaQiao.xlsx.In addition, the table locationw_ith_dem.xlsx represents the corrected latitude and longitude coordinates obtained from the geometric model and DEM elevation data, indicating the position corresponding to the center point of the UAV image; The table pxGT_dem.xlsx is calculated based on the latitude and longitude positions and resolution of the satellite image, and the pixel coordinates of the center point of the UAV image on the satellite image.