Abstract: Thermal imaging plays a vital role in surveillance systems, specifically low-light or nighttime conditions. However, traditional methods have challenges in accomplishing accurate object recognition as well as skeleton generation in thermal images due to varying environmental conditions. To address these challenges and improve object recognition and skeleton generation accuracy in thermal images, the proposed You Only Look Once version 8 using Efficient net as backbone for Thermal Image (YOLO8-EFTI) algorithm was evaluated and compared with the traditional YOLO v8 to assess its effectiveness in recognizing objects and key points in thermal images. This study introduces a YOLO8-EFTI applied to thermal images, aimed at enhancing accuracy through performance parameters. It was hypothesized that the YOLO8-EFTI algorithm would show greater accuracy compared to state-of-the-art methods. The results of the experiments indicate that the proposed YOLO8-EFTI method enhances performance, achieving a 9.5% higher precision, 37.84% higher recall, 8.53% higher mAP@50, and a 7.95% higher mAP@50-95 than the traditional YOLOv8 architecture in object detection.

Keywords: Deep learning, Object detection, Surveillance system, Thermal imaging, YOLO algorithm

Pages: 212-223