Roadside LiDAR for Cooperative Safety Auditing at Urban Intersections: Toward Trustworthy V2X Infrastructure Intelligence

Bo Shang, Yiqiao Li

Keywords: roadside LiDAR, near-miss analysis, surrogate safety, cooperative perception, 3D detection, multi-object tracking

Abstract:

Urban intersections expose the limitations of single-vehicle perception pipelines, where occlusions, compressed reaction windows, and viewpoint constraints hinder reliable understanding of safety-critical interactions. In cooperative autonomous driving systems, roadside intelligence and Vehicle-to-Everything (V2X) communication offer a complementary sensing layer that extends observability beyond onboard perception.

This paper presents an end-to-end roadside LiDAR pipeline for infrastructure-assisted safety auditing at a geo-fenced urban intersection near the City College of New York. The system integrates 3D detection, multi-object tracking, registration-guided temporal refinement, and structured human-in-the-loop quality assurance to construct auditable trajectories for downstream surrogate safety analysis. Rather than treating perception as an isolated benchmark problem, we frame roadside sensing as an external audit reference for cooperative systems and large driving models operating under partial observability.

We analyze a human-labeled heavy vehicle–bicycle near-miss to compare direction-agnostic time-to-collision (TTC) with longitudinal TTC of the heavy vehicle. While TTC collapses below 1 s, longitudinal TTC remains above conservative braking thresholds, revealing that the interaction is driven by lateral intrusion and compressed reaction time rather than insufficient longitudinal stopping capacity. This mechanism-aware interpretation illustrates how infrastructure intelligence can provide interpretable supervision signals for cooperative perception and decision-making models.

Beyond the anchor case, automatically mined vehicle–VRU near-miss events in later unlabeled frames remain spatially concentrated in the same central conflict zone, suggesting persistent infrastructure-level risk patterns. We argue that roadside LiDAR, when embedded within V2X architectures, can serve as a scalable foundation for trustworthy cooperative autonomy by enabling auditable event mining, exposure-aware safety evaluation, and calibration of multi-agent driving systems in complex urban environments.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 16