The study proposes a method for evaluating the significance of roadside billboards through videos captured from a driver’s perspective. A new dataset, BillboardLamac, was collected and annotated, consisting of eight videos captured by drivers equipped with eye-tracking devices. Various object tracking methods were evaluated in combination with a YOLOv8 detector to identify billboard advertisements. A random forest classifier was trained to classify billboards into three classes based on the length of driver fixations, achieving 75.8% test accuracy. The study reveals the duration of billboard visibility, its saliency, and size as the most influential features when assessing billboard significance.
Publication date: 14 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.07390
