Overcoming_data_scarcity_in_roadside_thermal_imagery

Published: 10 April 2025| Version 2 | DOI: 10.17632/66grzddyb2.2
Contributors:
Arnd Pettirsch, Alvaro García Hernandez

Description

Official dataset related to the paper: Overcoming data scarcity in roadside thermal imagery: A new dataset and weakly supervised incremental learning framework The paper is available at https://doi.org/10.3390/s25072340. Please cite the paper when using the dataset.

Files

Steps to reproduce

Two thermal image camera models the AXIS Q1942-E 10 mm and AXIS Q1952-E 10 mm were used to collect the data from the roadside. Both sensors record videos with 30 FPS and have 640 x 480 pixels resolution. The images are decoded in 8-bit format, where 255 is assigned to the hottest and 0 to the coldest pixel. These cameras were installed at 21 locations. These 21 locations include 12 inner-city spots, 5 locations on rural roads and 4 on the highway. The open-source tool LabelImg was used to create 2d bounding box annotations and class labels for every object of the classes: motorcyclist, car, bus, truck, pedestrian, bicyclist or e-scooter

Institutions

  • Rheinisch Westfalische Technische Hochschule Aachen

Categories

Object Detection, Thermal Imaging

Licence