ITSC 2025 Paper Abstract

Paper VP-VP.118

Ni, Ying (Tongji University), Li, Siying (Tongji University), Fan, Jialin (Tongji University)

VRU-Centric Hazardous Scenario Detection Via Monocular Spatiotemporal Feature Fusion

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Keywords Real-time Incident Detection and Emergency Management Systems in ITS, Safety Verification and Validation Methods for Autonomous Vehicle Technologies, Deep Learning for Scene Understanding and Semantic Segmentation in Autonomous Vehicles

Abstract

The rapid deployment of autonomous vehicles necessitates robust hazard detection systems, yet existing methods struggle to detect subtle spatiotemporal cues in hazardous vehicle-vulnerable road user (VRU) interactions. This paper addresses this critical gap through three key contributions. First, a multi-stream spatiotemporal network is proposed to synergistically integrates 3D bounding box localization and optical flow dynamics via a transformer-based fusion architecture. The framework extracts geometric context from 3D object detection, encodes motion patterns through optical flow estimation, and employs temporal self-attention to model hazard precursors. Second, we introduce VRUHI dataset—a benchmark tailored for vehicle-VRU interactive risk assessment, comprising 6,000 dashcam clips spanning diverse urban scenarios. Third, extensive experiments demonstrate state-of-the-art performance. Our model outperforms prior methods and validating its efficacy in capturing early hazardous signals. The dataset and model establish a foundation for safety-critical autonomous systems to detect VRU-related hazards proactively.