Thermal Image Dataset

About Dataset

ABSTRACT

The Thermal Vision dataset is a valuable resource for training machine learning models to detect illegal movements around borders and in protected areas.

The dataset comprises 7412 manually labeled images extracted from video frames captured in the long-wave infrared (LWIR) spectrum. 

Instructions:

In terms of processing, approximately 20 minutes of recorded material from clear weather scenarios, 13 minutes from fog scenarios, and about 15 minutes from rainy weather scenarios were utilized. Longer videos were segmented into sequences, and individual frames were extracted, resulting in 11,900 images for clear weather, 4,905 images for fog, and 7,030 images for rainy weather scenarios.

These annotated frames were selected to include various weather conditions, with 2,663 frames in clear weather, 1,135 frames in fog, and 2,313 frames in rain scenarios.

The annotations were performed using the Yolo Box Annotation Tool, which supports storing annotations in YOLO, VOC, and MS COCO formats. Each annotation includes the centroid position of the bounding box around each object, the bounding box’s width and height, and the corresponding class label (Human or Dog).