CNN-Based Ball and Goal Detection for KRSBI Robot with Omnidirectional Camera

The Wheeled Soccer Robot Contest (KRSBI-Beroda) challenges robots to autonomously detect, dribble, and score using vision-based systems. Traditional object detection methods like HSV color filtering are widely used but perform poorly under varying lighting conditions. This study proposes a Convolutional Neural Network (CNN)-based object detection system using the YOLO (You Only Look Once) algorithm to enhance the accuracy and reliability of ball and goal detection in KRSBI robots equipped with omnidirectional cameras. A dataset of 1,125 images comprising diverse lighting and object positions was collected and split into 80% training and 20% validation sets. The YOLOv8 model was trained using Ultralytics on Google Colab with 100 epochs. The resulting model achieved a high detection performance, with an accuracy of 95.87%, precision of 1.00 at a confidence threshold of 0.921, recall of 0.99, and an F1-Score of 0.97. The results confirm that the YOLOv8-based CNN model provides a robust and efficient solution for real-time ball and goal detection in robotic soccer applications.