Pot Detection System Using YOLO

The ABU Robocon 2021 contest is a contest of throwing arrows into a pot using a robot. The highlight of this game is how the robot can throw arrows into the pot so that it gets a score within a predetermined time limit. In this study, a pot detection system is needed to know the pot to be detected and the estimated distance from the robot to the pot. At this time Deep Learning is often used for research and finding new methods such as YOLO (You Only Look Once). YOLO architecture requires a sophisticated computer architecture to run, so the solution is YOLOv3-tiny-prn. YOLOv3-tiny-prn is a more compact architecture than YOLOv3 and YOLOv3-tiny because the depth of convolutional layers is reduced, so the detection process is faster. So the method used in this study for pot detection is the YOLOv3-tiny-prn method. The error value obtained from the results of the distance estimation test using the YOLOv3-tiny-prn method is below 5%, but in the Scoring Spot 4 distance test there is an error of 100
^%
because the pot is covered by the pot in front of it so it cannot be detected.