| Title |
A Study on the Object Detection Algorithm based on Deep Learning for Port Monitoring System |
| Authors |
이유진(Yujin Lee) ; 이순교(Soonkyo Lee) ; 최계원(Kae Won Choi) ; 박재현(Jaehyun Park) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.2.39 |
| Keywords |
Deep learning; Object detection; Monitoring system; Port security; Smart port |
| Abstract |
Since 2008, an average of 25 security accidents have been steadily occurring at ports, which are national security facilities. In existing ports, technologies have been introduced to prevent security accidents such as smuggling using CCTV cameras or sensors, but false warnings often occur due to weather or illuminance changes. In addition, existing technologies cannot manage outsiders' access to dangerous areas when they enter the port. Therefore, in this paper, we propose a new port monitoring system that sends notifications to managers by determining that a specific object is a dangerous situation when approaching a dangerous area through object detection using deep learning, and to this end, we studied a deep learning-based object detection algorithm. A learning data set was built by collecting CCTV images of the Busan Port terminal, and model learning was conducted using deep learning-based object detection algorithms YOLOv3 and CenterNet. Comparing the detection performance of the two models, it was confirmed that the performance of the YOLov3 model was 94.1%, 33.7% higher than that of the CenterNet model, and that the object detection model suitable for the port environment was YOLov3. As a result of experimenting with the YOLov3 model using actual port images, it was confirmed that object detection was successfully performed in various places, times, and weather. |