A post-disaster survivor dataset is used for UAV post-disaster personnel search and rescue and statistics, it contains the real-world post-disaster UAV images from multiple perspectives, scenes, and distances.
File "879real survivor" is a collection of survivors' images of post-disaster ruins for UAV search and rescue. It contains 879 images and corresponding "label" files.
File "155real survivor" is the test dataset from reference [1][2], containing four email screenshots. In this study, 155 images out of 250 images and corresponding "label" files were applied. Due to copyright issues, it is not convenient to publish this dataset. And if necessary, you can contact the author of the paper to obtain this dataset.
File "112TinyPerson" is the corresponding post-disaster crowd image and the corresponding "label" file.
The "label" files are the tex file in YOLO format.
[1] Zhang N, Nex F, Vosselman G, et al. Unsupervised harmonious image composition for disaster victim detection[J]. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2022, 43: 1189-1196.
[2] Zhang N, Nex F, Vosselman G, et al. Training a disaster victim detection network for UAV search and rescue using harmonious composite images[J]. Remote Sensing, 2022, 14(13): 2977.
The models and codes are available. Click here