WiSARD
Labeled visual and thermal images of various wilderness environments

WiSARD stands for Wilderness Search and Rescue Dataset (pronounced "wizard"). WiSARD consists of visual and thermal imagery taken from a drone flying over various wilderness environments in Washington, USA. The purpose of the WiSAR Image Dataset is to advance computer vision and deep learning research with a targeted application for wilderness search and rescue.

The dataset is collected by students from the Autonomous Flight Systems Lab located in the Aeronautics and Astronautics Department at the University of Washington.

What is WiSAR?

Wilderness Search and Rescue (WiSAR) is the process of finding, geolocating, assisting and evacuating person(s) lost in wilderness areas. Wilderness areas are typically characterized by rugged, difficult-to-access terrain and dense vegetation cover. These factors not only make the task of locating persons difficult and slow, but also represent a safety risk to the search team who must navigate the hazardous terrain on-foot. Moreover, WiSAR operation expenses are overwhelmingly dominated by personnel costs, attributing to over 80% of the overall expense. The cost varies from incident to incident, but for rugged terrain such as the Enchantment Range in Washington State, the typical cost is between $8,500 to $10,500.

Why WiSARD?

Given the challenges that WiSAR crews face, there is a clear need to develop solutions that can (i) reduce search times since time is of the essence, (ii) reduce the overall mission cost, and (iii) reduce safety risk to the WiSAR personnel. We believe that using camera-equipped Unoccupied Aerial Vehicles (UAVs) to aid WiSAR operations can effectively address all the aforementioned needs. With recent advances in on-board sensors and computation, the solution would entail running object (specifically, human) detection algorithms onboard the UAV and the information be disseminated to the WiSAR team in real-time.

The goal of the WiSARD is to provide a benchmark for computer vision research that benefit WiSAR applications and alike (e.g., search and rescue in natural disaster zones). Existing dataset benchmarks do not contain enough wilderness-specific examples with "lost" humans. A new dataset benchmark is necessary due to the unique settings of WiSAR operations; the presence of unstructured wilderness terrain, occlusion due to foliage, size and sparsity of human objects, and diversity in weather and lighting conditions (e.g., snow, fog, forests, mountains, dawn, dusk, night).

To the best of our knowledge, WiSARD is the first large-scale wilderness-specific image dataset that consists of multimodal data (visual and thermal images) taken in a variety of different wilderness terrains in many different weather and lighting conditions.

What is in the WiSAR Image Dataset?

The dataset includes three main subsets of data, the "Visual Only" set of 26,862 labeled visual images, the "Thermal Only" set of 29,989 labeled Long Wave Infrared (thermal) images, and the "Multi-modal" set, a subset of the Visual and Thermal Only sets consisting of 15,453 temporally synchronized (i.e., taken at the same instant in time) visual-thermal image pairs.

See our paper for more details about the data collection, labeling, and annotation processes.

Download the WiSAR Image Dataset

WiSARDv1: a large (40.54 GB) file containing Visual Only, Thermal Only, and Multi-Modal datasets. | [Download link]

WiSARD Multi Modal Sample: a smaller (971.6 MB) file of synchronized and annotated visual-thermal image pairs from a single data collection flight. | [Download link]

People

Danny Broyles
University of Washington
Graduate student

Chris Hayner
University of Washington
Graduate student

Kenny Wilsey
University of Washington
Undergraduate student

Annika Singh
High school intern

Karen Leung
University of Washington
Assistant Professor

Juris Vagners
University of Washington
Emeritus Professor



Contact

Send questions and enquiries to afsl@uw.edu

Citation

Please use the following citation when referencing this dataset and associated work.

@inproceedings{BroylesHaynerEtAl2022,
author = {Broyles, D.* and Hayner, C.* and Leung, K.},
booktitle = {{IEEE/RSJ Int.\ Conf.\ on Intelligent Robots \& Systems}},
title = {{WiSARD}: A Labeled Visual and Thermal Image Dataset for Wilderness Search and Rescue},
year = {2022},

}


Acknowledgements

We would like to acknowledge Will Browne, Evelyn Madwell, Timothy Zhou, Neil Gupta, Helen Kuni, and Karina Bridgman for their help in collecting and annotating the data.

License

Copyright 2022 Daniel Broyles, Christopher Hayner, Karen Leung

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.