Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VRU-CIPI: Crossing Intention Prediction at Intersections for Improving Vulnerable Road Users Safety

Ahmed S. Abdelrahman, Mohamed Abdel-Aty, Quoc Dai Tran

TL;DR

This work tackles the challenge of predicting VRU crossing intentions at urban intersections to prevent conflicts with vehicles. It introduces VRU-CIPI, a real-time framework that fuses GRU-based temporal dynamics with a multi-head Transformer to integrate pose, motion, and geometric cues from CCTV camera data. On the diverse UCF-VRU dataset, VRU-CIPI achieves state-of-the-art accuracy (96.45%) with real-time inference at 33 FPS and demonstrates potential for proactive I2V communication and intersection signal management. The study highlights the critical role of pose estimation in disambiguating crossing directions, especially in complex waiting areas, and provides a practical path toward safer, smoother multi-modal urban mobility.

Abstract

Understanding and predicting human behavior in-thewild, particularly at urban intersections, remains crucial for enhancing interaction safety between road users. Among the most critical behaviors are crossing intentions of Vulnerable Road Users (VRUs), where misinterpretation may result in dangerous conflicts with oncoming vehicles. In this work, we propose the VRU-CIPI framework with a sequential attention-based model designed to predict VRU crossing intentions at intersections. VRU-CIPI employs Gated Recurrent Unit (GRU) to capture temporal dynamics in VRU movements, combined with a multi-head Transformer self-attention mechanism to encode contextual and spatial dependencies critical for predicting crossing direction. Evaluated on UCF-VRU dataset, our proposed achieves state-of-the-art performance with an accuracy of 96.45% and achieving real-time inference speed reaching 33 frames per second. Furthermore, by integrating with Infrastructure-to-Vehicles (I2V) communication, our approach can proactively enhance intersection safety through timely activation of crossing signals and providing early warnings to connected vehicles, ensuring smoother and safer interactions for all road users.

VRU-CIPI: Crossing Intention Prediction at Intersections for Improving Vulnerable Road Users Safety

TL;DR

This work tackles the challenge of predicting VRU crossing intentions at urban intersections to prevent conflicts with vehicles. It introduces VRU-CIPI, a real-time framework that fuses GRU-based temporal dynamics with a multi-head Transformer to integrate pose, motion, and geometric cues from CCTV camera data. On the diverse UCF-VRU dataset, VRU-CIPI achieves state-of-the-art accuracy (96.45%) with real-time inference at 33 FPS and demonstrates potential for proactive I2V communication and intersection signal management. The study highlights the critical role of pose estimation in disambiguating crossing directions, especially in complex waiting areas, and provides a practical path toward safer, smoother multi-modal urban mobility.

Abstract

Understanding and predicting human behavior in-thewild, particularly at urban intersections, remains crucial for enhancing interaction safety between road users. Among the most critical behaviors are crossing intentions of Vulnerable Road Users (VRUs), where misinterpretation may result in dangerous conflicts with oncoming vehicles. In this work, we propose the VRU-CIPI framework with a sequential attention-based model designed to predict VRU crossing intentions at intersections. VRU-CIPI employs Gated Recurrent Unit (GRU) to capture temporal dynamics in VRU movements, combined with a multi-head Transformer self-attention mechanism to encode contextual and spatial dependencies critical for predicting crossing direction. Evaluated on UCF-VRU dataset, our proposed achieves state-of-the-art performance with an accuracy of 96.45% and achieving real-time inference speed reaching 33 frames per second. Furthermore, by integrating with Infrastructure-to-Vehicles (I2V) communication, our approach can proactively enhance intersection safety through timely activation of crossing signals and providing early warnings to connected vehicles, ensuring smoother and safer interactions for all road users.
Paper Structure (18 sections, 11 equations, 3 figures, 4 tables)

This paper contains 18 sections, 11 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Crossing Intention Prediction from Vehicle Camera View
  4. Crossing Intention Prediction at Intersections
  5. Methodology
  6. VRU Crossing Prediction and Crossing Monitoring Framework
  7. First Stage | ZoI Declaration
  8. Second Stage | Video Processing and Feature Extraction
  9. Third Stage | Crossing Prediction
  10. Fourth Stage | Crossing Monitoring
  11. VRU-CIPI: VRU Crossing Intention Prediction at Intersections
  12. Experiemnts
  13. UCF-VRU Dataset
  14. Evaluation Metrics
  15. Experimental Setting
  16. ...and 3 more sections

Figures (3)

  • Figure 1: VRU-CIPI framework for VRU monitoring and crossing prediction at intersections. It integrates detection, tracking, and pose estimation models to extract individual features while considering the geometric design of the waiting area for accurate prediction.
  • Figure 2: VRU-CIPI Model Architecture for Crossing Intention Prediction. Features are filtered, concatenated, and then passed to GRU to capture temporal dependencies, while multi-head self-attention transformer encoder models contextual relationships. The final prediction is made through fully connected layers with a sigmoid activation function. TF: temporal filtering. C: concatenation. FC: Fully Connected Layer.
  • Figure 3: The qualitative results of VRU-CIPI. Examples illustrating model predictions for different VRU types and under different weather and lighting conditions. Cases include: (a) a scooter rider during clear daytime conditions; (b) a scenario with seven pedestrians; and (c) a cyclist during nighttime with rainy weather.