Real-Time Object Recognition for Football Field Landmark Detection Based on Deep Neural Networks

摘要

The main goal of this study is to detect the landmark of football fields with a real-time object recognition system based on deep neural networks. The deep neural networks used are YOLO (You Only Look Once). The deep learning of YOLO can detect and recognize objects robustly in a high speed. By using this method, the robot can robustly detect and recognize the landmark of football fields in images with size of 608 × 608 pixels at 16 frames per second with MAP (Mean Average Precision) and IoU (Intersect over Union) with 97.60% and 81.36% accuracy.