RE-IDENTIFIKASI ORANG PADA SISTEM PEMANTAUAN KELUAR-MASUK KAMERA PENGAWAS TUNGGAL
Muhammad Nur Ilmi, Prof. Drs. Agus Harjoko, M.Sc., Ph.D; Ika Candradewi, S.Si., M.Cs.
2023 | Skripsi | ELEKTRONIKA DAN INSTRUMENTASI
Entry-Exit Monitoring System is a part of intelligent surveillance systems. This issue aims to monitor persons' activity using wide-ranging surveillance cameras in both public and private spaces, such as restrooms, changing rooms, and baby care areas.
This research presents the development and evaluation of a single-camera entry-exit monitoring system equipped with person re-identification to maintain a consistent identity while people going in and out of an area. The system employs, comprehensive pipeline, including YOLOv7 (original and tiny versions), ByteTrack, and OSNet-x1 as the person detector, tracker, and re-identifier.
The entry-exit event detection component using the YOLOv7+ByteTrack achieved the highest accuracy, producing a value of 1 for both entry and exit event detection. Conversely, the YOLOv7-tiny+ByteTrack system showed slightly lower performance with an entry event detection score of 0,975.
The best person re-identification performance was observed with YOLOv7+ByteTrack+OSNet-x1, achieving an f1-score of 0,864 and accuracy of 0,826, whereas YOLOv7-tiny+ByteTrack+OSNet-x1 exhibited reduced values of 0,826 and 0,780, respectively. FPS evaluation results on Google Colab and Nvidia Jetson Xavier AGX show that the tiny model YOLOv7+ByteTrack+OSNet-x1 achieved the highest average FPS, with 26,7 FPS on Google Colab and 13 FPS on Nvidia Jetson Xavier AGX. In contrast, when using the YOLOv7+ByteTrack+OSNet-x1 model, performance was slightly lower, with average FPS of 21,8 FPS and 10,6 FPS on the respective platforms.
Kata Kunci : Sistem pemantauan keluar-masuk orang, Kamera pengawas tunggal, Intelligent surveillance system, YOLOv7, ByteTrack, OSNet-x1