| Title |
Evaluation Methodology of Low-Light Image Enhancement Methods for Improving Real-Time Human Detection Performance of Indoor Night Patrol Robot |
| Authors |
이재영(Lee, Jae Young) ; 김수민(Kim, Soo Min) ; 홍성철(Hong, Sung Chul) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.2.0277 |
| Keywords |
실내 야간 순찰 로봇; 저조도 영상강화; YOLOv8n-seg; 사람 탐지 Indoor night patrol robot; Low-light image enhancement; YOLOv8n-seg; Human detection |
| Abstract |
Various types of patrol robots have been introduced to overcome the limitations of indoor security and crime prevention caused by shortages of security personnel and blind spots in CCTV systems. The optical images taken by patrol robot support remote operators in controlling the robot, understanding the patrol area, and making decisions. When combined with deep learning-based object detection methods, these images enable the rapid and accurate identification of interest objects. Particularly, human detection by patrol robots is an essential function for intruder detection, safety monitoring, and incident detection. However, patrol images captured in low-light indoor environments during nighttime are dark and noisy, making it difficult for patrol robots to perform their tasks effectively. In this study, low-light image enhancement methods (GLADNet, KinD, TBEFN, LLFormer, EnlightenGAN, Zero-DCE) were applied to nighttime indoor patrol images to analyze their effectiveness in improving visibility and enhancing human detection performance using YOLOv8n-seg model. The results showed that the color and brightness of low-light images were effectively restored in the KinD, TBEFN, and LLFormer images, leading to improved visibility and significantly enhanced image quality metrics. Also, the human detection accuracy of the YOLOv8n-seg model increased in the order of KinD and LLFormer images. KinD enhanced the real-time visibility of nighttime patrol images and significantly improved human detection performance. These findings are expected to increase the performance of night patrol robot using optical cameras. |