Computer vision for package tracking on omnidirectional wheeled conveyor: Case study

In this paper, a real-time camera tracking system for package transportation on omnidirectional wheeled conveyor is presented. The camera tracking system is integrated with a closed-loop controller for packages path planning. No additional sensors are used for the controller implementation, only a 2-D Camera. The package’s position and orientation are detected by the camera tracking system in real-time. Two proposed systems are presented, System I is implemented using the conventional image processing technique threshold method while System II is implemented using computer vision. In System II, three computer vision models were evaluated: Detectron2, YOLOv5 and Faster R-CNN. Experimental results in real-time show that System I have lower accuracy rate 85.7% compared to System II which reported 98% and 88.1% for YOLOv5 and Detectron2, respectively. YOLOv5 reported the best results among the computer vision models with 1% missing rate, 45.5 FPS and average precision of 99.8%.