Yao, Ziying (Beihang University), Xiong, Zhongxia (Beihang University), Liu, Xuan (Beihang University), Wu, Xinkai (Beihang University)

TopoTP: Augmenting Driving Topology Reasoning with Dynamic Traffic Participants

Scheduled for presentation during the Poster Session "Detection, estimatation and prediction for intelligent transportation systems" (WeAT17), Wednesday, September 25, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Topology reasoning in autonomous driving focuses on thoroughly analyzing traffic environments to identify feasible driving paths. This challenging task involves identifying lanes and traffic elements, then deducing the relationships between lanes (lane-lane topology), lanes and traffic elements (lane-traffic topology). It is a challenging task due to the dynamic and complex nature of traffic environments, and the common issue of visual obstructions. In this paper, we propose TopoTP, a high-performance end-to-end model for driving topology reasoning considering dynamic traffic participants. We introduce traffic participants decoder module into the united framework and integrate informative dynamic clues implicitly with static features from cross space, enabling a deeper level of traffic scene analysis in complex environments. TopoTP achieves state-of-the-art performance on OpenLane-V2 benchmark, with results showcasing its capability to deliver reliable topology reasoning in complicated and dynamic driving scenarios.