HIT-UAV: A high-altitude infrared thermal dataset for Unmanned Aerial Vehicle-based object detection

Introduction

The HIT-UAV contains 2898 infrared thermal images extracted from 43470 frames, captured by UAV from different scenes (schools, parking lots, roads, playgrounds, etc.), covering a wide range of aspects including objects (Person, Bicycle, Car, OtherVehicle), flight altitude data (from 60 to 130 meters), camera perspective data (from 30 to 90 degrees), and daylight intensity (day and night).

Annotation

The HIT-UAV provide two bounding box type, oriented and standard. The oriented annotation can decrease the overlap of bounding boxes to improve the performance of detection algorithms.

• Standard bounding box record format: $[xc, yc, w, h]$.
• Oriented bounding box record format: $[xc, yc, w, h, \theta]$, where $\theta$ denotes the oriented angle from the horizontal direction of the standard bounding box.

For each annotation method, we provide the XML and JSON label file to help user utilize the HIT-UAV:

• normal_xml folder: record standard bounding boxes using xml file.
• normal_json folder: record standard bounding boxes using json file.
• rotate_xml folder: record oriented bounding boxes using xml file.
• rotate_json folder: record oriented bounding boxes using json file.

Sample images

The detection samples using YOLOv4.

Training Examples

YOLOv8: https://www.kaggle.com/code/binh234/yolov8-training-on-hit-uav

Paper

https://doi.org/10.1038/s41597-023-02066-6

Cite

Jiashun Suo, Tianyi Wang, Xingzhou Zhang, Haiyang Chen, Wei Zhou and Weisong Shi. HIT-UAV: A high-altitude infrared thermal dataset for Unmanned Aerial Vehicle-based object detection. Scientific Data 10, 227 (2023).

or

@article{suo2023hit,  
    title       = {HIT-UAV: A high-altitude infrared thermal dataset for Unmanned Aerial Vehicle-based object detection},  
    author      = {Suo, Jiashun and Wang, Tianyi and Zhang, Xingzhou and Chen, Haiyang and Zhou, Wei and Shi, Weisong},  
    journal     = {Scientific Data},  
    volume      = {10},
    pages       = {227},
    year        = {2023},
    publisher   = {Nature Publishing Group UK London}
}