A Foveated Approach to Automated Billboard Detection

Understanding billboard visibility is vital when considering the value of each billboard to advertisers, hence the growing demand for artificial intelligence based approaches to visibility measurement. Addressing this need, this research
paper presents a comprehensive approach to billboard detection using street-view images. We have developed a robust billboard detection system by leveraging state-of-the-art object detection models, such as You Only Look Once (YOLOv8), YOLOv5, Faster-Region-based Convolutional Neural Network (RCNN) and CenterNet resulting in high model accuracy. We have introduced an innovative foveated approach, based on the human visual systems, that applies a Gaussian function to assign weights to billboards to determine which is the most significant billboard based on a combination of confidence and location with respect to the image centre. The approach demonstrates an improvement in overall accuracy of the detection process. In particular YOLOv8 experienced a high accuracy increase from 63.40 to 82.71 percent. This research provides valuable insights and practical solutions for billboard detection in real-time.