U.S. flag An official website of the United States government.
Official websites use .gov

A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS

A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

i

Investigation of Emerging Sensing and AI/ML Technologies to Enhance the Safety of Vulnerable Roadway Users at Signalized Intersection

File Language:
English


Select the Download button to view the document
Please click the download button to view the document.

Details

  • Creators:
  • Corporate Creators:
    City College of New York
  • Corporate Contributors:
  • Subject/TRT Terms:
  • Resource Type:
    Tech Report
  • Geographical Coverage:
    United States
  • Edition:
    Final Report: 2024–2026
  • Corporate Publisher:
    United States. Department of Transportation. University Transportation Centers (UTC) Program
  • Abstract:
    Vulnerable roadway users (VRUs)—including pedestrians, cyclists, and micromobility users—face increasing safety risks at signalized intersections. However, traditional crash-based monitoring approaches are limited in their ability to capture near-miss events and behavioral conflicts that precede crashes. This report investigates how emerging LiDAR-based sensing and AI/ML technologies can improve VRU safety observability at signalized intersections through three interconnected research thrusts. First, a comprehensive survey of infrastructure-based sensing modalities and AI/ML methods is presented to examine the capabilities, limitations, and challenges of existing approaches. Future research directions are identified to improve the robustness, scalability, and effectiveness of AI-powered intersection safety systems. Among the reviewed sensing technologies, LiDAR emerges as a particularly promising primary modality for trajectory-level safety analysis due to its geometric precision and robustness under varying lighting conditions. Second, to evaluate the reliability of LiDAR-based VRU detection under sensor degradation, a systematic study of vertical LiDAR beam loss is conducted across six 3D detection architectures using the KITTI (64-beam) and nuScenes (32-beam) benchmark datasets. The results show that VRU detection performance remains relatively stable up to approximately 20% beam loss but deteriorates rapidly beyond this threshold. Furthermore, contiguous beam loss—commonly caused by sensor occlusion or lens contamination—is found to be substantially more detrimental than dispersed beam loss of equivalent magnitude. These findings provide actionable insights for sensor maintenance, model selection, and operational risk assessment in infrastructure-based sensing systems. Third, a roadside LiDAR-camera data collection site was established at a signalized intersection in New York City, including object-level annotations for pedestrians and cyclists. Building upon this dataset, an end-to-end auditable safety-analysis framework is developed and demonstrated using an 8,000-frame manually annotated roadside LiDAR dataset. The framework integrates 3D detection, multi-object tracking, trajectory refinement, dynamics-aware stabilization, and structured human-in-the-loop quality assurance to transform raw sensor observations into defensible near-miss safety evidence. A case study involving a heavy vehicle-bicycle interaction demonstrates how combined direction-agnostic and longitudinal time-to-collision analyses can reveal lateral-intrusion-dominated conflict mechanisms that single-metric approaches fail to distinguish. Together, these three contributions establish roadside LiDAR as a viable foundation for scalable, interpretable, and auditable VRU safety monitoring at signalized intersections.
  • Format:
    PDF
  • Funding:
  • Collection(s):
  • Main Document Checksum:
    urn:sha-512:b41f3bb79b95825c1f4b793442ac7c7b812de8e0abaea9c5415cb01f22552fc385a7f12c163ef62eaaa539f9d8c41c8cf699dd133296c7360cf819ee08540e19
  • Download URL:
  • File Type:
    Filetype[PDF - 13.69 MB ]
PREVIOUS
File Language:
English
ON THIS PAGE
Details
ROSA P serves as an archival repository of USDOT-published products including scientific findings, journal articles, guidelines, recommendations, or other information authored or co-authored by USDOT or funded partners. As a repository, ROSA P retains documents in their original published format to ensure public access to scientific information.
BACK TO TOP