• Zou. Z., Zhang R., Shen, S., Pandey G., Chakravarty P., Parchami A., and Liu, H. X. (2022). Real-time Full-stack Traffic Scene Perception for Autonomous Driving with Roadside Cameras. Accepted by the 2022 IEEE International Conference on Robotics and Automation (ICRA), May 23-27, 2022, Philadelphia, USA.
    DOI: 10.48550/arXiv.2206.09770
  • Zhang R., Zou Z., Shen S., Liu H.X. (2022). Design, Implementation, and Evaluation of a Roadside Cooperative Perception System. Transportation Research Record. June 2022.
    DOI: 10.1177/03611981221092402


  • Meng D., Sayer O., Zhang R., Shen S., Li H., and Liu H. X. (2023). ROCO: A Roundabout Traffic Conflict Dataset, presented at TRB 2023.
    Slide Deck
  • Guo H., Feng Y., Shen S., Zhang R., Liu H.X., Bao S., (2023). An Examination of Driver Behavior at a Roundabout Using Trajectory Data from a Roadside Cooperative Perception System, presented at TRB 2023.
    Slide Deck

Cooperative Driving Automation

MSight: Infrastructure-based Perception System for Cooperative Driving Automation

  • A full-stack cooperative perception system that closes the loop between roadside perception and V2X communication for cooperative driving automation
  • The algorithm is deep learning (DL) based and offers state-of-the-art performance with high accuracy and low latency
  • Production-ready edge-cloud infrastructure for large-scale deployment
  • The roundabout trajectory dataset contains the vehicle trajectory data perceived by the MSight roadside perception system deployed at the two-lane roundabout at the intersection of State St. and W. Ellsworth Rd. in Ann Arbor, Michigan.
Program view of the State and Ellsworth Roundabout with vehicles and their trajectory data.

MSight: Traffic Conflict & Crash Detection

  • A learning-based conflict identification algorithm from video collected at the two-lane roundabout at the intersection of State St. and W. Ellsworth Rd. in Ann Arbor, Michigan.
  • Provide guidance to traffic agency for crash-prone location identification and crash prevention
  • EEnable naturalistic driving environment building
  • The ROudanbout traffic COnflict (ROCO) dataset is a collection of traffic conflict events. Each event captures a 30-second duration of the conflict. The dataset provides the trajectories of the conflicts, along with information about the reason, time, and effect of conflict.