Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Billboard in Focus: Estimating Driver Gaze Duration from a Single Image

Carlos Pizarroso, Zuzana Berger Haladová, Zuzana Černeková, Viktor Kocur

TL;DR

This work tackles driver gaze toward roadside billboards and the associated distraction risk by proposing a fully automated, two-stage pipeline. A YOLO-based billboard detector is pre-trained on Mapillary Vistas and fine-tuned on BillboardLamac, followed by a gaze-duration classifier that uses bounding-box geometry and DINOv2 features to estimate gaze from individual frames. The method achieves 94% mAP@50 in billboard detection and 68.1% per-frame accuracy on BillboardLamac, with further validation on Google Street View achieving 66.3% accuracy. While offering a scalable, single-image alternative to eye-tracking-based approaches, the study notes limitations due to the small number of unique billboards and suggests future work to collect larger, more diverse data; code and data are publicly available.

Abstract

Roadside billboards represent a central element of outdoor advertising, yet their presence may contribute to driver distraction and accident risk. This study introduces a fully automated pipeline for billboard detection and driver gaze duration estimation, aiming to evaluate billboard relevance without reliance on manual annotations or eye-tracking devices. Our pipeline operates in two stages: (1) a YOLO-based object detection model trained on Mapillary Vistas and fine-tuned on BillboardLamac images achieved 94% mAP@50 in the billboard detection task (2) a classifier based on the detected bounding box positions and DINOv2 features. The proposed pipeline enables estimation of billboard driver gaze duration from individual frames. We show that our method is able to achieve 68.1% accuracy on BillboardLamac when considering individual frames. These results are further validated using images collected from Google Street View.

Billboard in Focus: Estimating Driver Gaze Duration from a Single Image

TL;DR

This work tackles driver gaze toward roadside billboards and the associated distraction risk by proposing a fully automated, two-stage pipeline. A YOLO-based billboard detector is pre-trained on Mapillary Vistas and fine-tuned on BillboardLamac, followed by a gaze-duration classifier that uses bounding-box geometry and DINOv2 features to estimate gaze from individual frames. The method achieves 94% mAP@50 in billboard detection and 68.1% per-frame accuracy on BillboardLamac, with further validation on Google Street View achieving 66.3% accuracy. While offering a scalable, single-image alternative to eye-tracking-based approaches, the study notes limitations due to the small number of unique billboards and suggests future work to collect larger, more diverse data; code and data are publicly available.

Abstract

Roadside billboards represent a central element of outdoor advertising, yet their presence may contribute to driver distraction and accident risk. This study introduces a fully automated pipeline for billboard detection and driver gaze duration estimation, aiming to evaluate billboard relevance without reliance on manual annotations or eye-tracking devices. Our pipeline operates in two stages: (1) a YOLO-based object detection model trained on Mapillary Vistas and fine-tuned on BillboardLamac images achieved 94% mAP@50 in the billboard detection task (2) a classifier based on the detected bounding box positions and DINOv2 features. The proposed pipeline enables estimation of billboard driver gaze duration from individual frames. We show that our method is able to achieve 68.1% accuracy on BillboardLamac when considering individual frames. These results are further validated using images collected from Google Street View.
Paper Structure (13 sections, 3 figures, 3 tables)

This paper contains 13 sections, 3 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Object Detection for Billboard Analysis
  4. Datasets for Billboard and Scene Understanding
  5. Automation in Gaze and Attention Estimation
  6. DATA
  7. PROPOSED METHOD
  8. Billboard Detection
  9. Billboard Detection Results
  10. Gaze Duration Estimation
  11. RESULTS
  12. Evaluation Using Google Street View
  13. CONCLUSION

Figures (3)

  • Figure 1: Example of billboard detections using the fine-tuned YOLO11 model.
  • Figure 2: Example of classified billboards into gaze duration categories.
  • Figure 3: Billboard from the BillboardLamac test set (top) and its corresponding Google Street View example in different views (bottom).