A Real-Time Carboy Tracking and Counting System Based on YOLOv8 and ByteTrack

In industrial processes, the accurate detection, counting, and classification of objects are crucial for the efficiency of production lines. However, when these processes are performed manually, human errors and time loss can occur. In this study, a YOLOv8-based system was developed to separate carboys based on their expiration dates, count solid carboys, and detect carboys containing foreign matter at the Pürsu Water Factory. The system, which processes real-time video streams, integrates YOLOv8 for object detection, tracking, and counting. Additionally, the performances of the DeepSORT and ByteTrack algorithms were compared for object tracking. The results show that ByteTrack performed faster and more stably than DeepSORT. The YOLOv8-based model achieved 98% accuracy, and ByteTrack worked at a higher FPS than DeepSORT, thereby improving the efficiency of the production line. This study presents an in-depth analysis of the system’s accuracy, speed, and overall performance.