Deploy Intersection Roadside Computers that Can Handle the Highly Demanding Pipeline for Real-time LiDAR Processing.
The Deployment of LiDAR Pedestrian Detection for Optimizing Signal Timings in Florida Offered Lessons on Implementing CAV Communications Technologies and Traffic Control Algorithms.
Made Public Date

Extended Development and Testing of Optimized Signal Control with Autonomous and Connected Vehicles


The goals of this study were to extend the scope of FDOT's Real-time Intersection Optimizer (RIO) to:

  1. incorporate pedestrians
  2. investigate how connected and automated vehicle (CAV) communications could enhance the safety and efficiency of CAVs at intersections
  3. use RIO for real-time anonymous tracking of vehicles and pedestrians; and
  4. implement a new optimization algorithm and evaluate its performance.

The research team designed and deployed a LiDAR-based pedestrian detection system into RIO, as well as pedestrian phasing capabilities. A simulated intersection which was modeled based on a real intersection in Gainesville, Florida was developed to evaluate how vehicle and pedestrian delays were impacted by RIO’s upgraded SigPT solver under different demand patterns for both pedestrians and vehicles when compared to an actuated signal. Field tests were also conducted at FDOT’s Traffic Engineering and Research Laboratory (TERL) to evaluate the hardware and software system.

Two images of a signalized intersection with a pedestrian crosswalk signal showing a "Do not Cross" signal (left image) and a "Cross" signal (right image)
Figure 1: LiDAR being used to call the ped phase. Source: FDOT

Lessons Learned

The lessons learned from the deployment of the RIO system were from development experiences, and the performance of the system with respect to the project goals and effects on vehicular and pedestrian delays at intersections.

  • Rotate to the camera-centric world coordinate system, as needed. After connecting to the Real Time Streaming Protocol (RTSP) stream, the team noticed that the camera was tilted such that the road was rotated clockwise by about ~10 degrees, throwing off the video tracker. To address this, the team counteracted this by applying a ~10-degree rotation counter-clockwise using image processing techniques before detecting the vehicles.
  • Use a fast roadside unit (RSU) computer for running optimization algorithms. In this study, the RSU computer was pushed to its limit even though the total traffic demand was very small. A fast processor would be recommended for running optimization algorithms like RIO for the system to be able to detect and optimize higher traffic demands. This will allow optimization algorithms to run at higher frequencies to react more quickly to changing traffic conditions.
  • Install additional/stronger LiDAR to detect pedestrians within the entire intersection. In this study, the range of the LiDAR was insufficient to detect the opposite corners of the intersection. Additional or a stronger LiDAR should be installed to observe the entire intersection. It is worth noting that a stronger LiDAR may require a higher position when implementing in the field.

Extended Development and Testing of Optimized Signal Control with Autonomous and Connected Vehicles

Extended Development and Testing of Optimized Signal Control with Autonomous and Connected Vehicles
Source Publication Date
Elefteriadou, Lily; Sanjay Ranka; Carl Crane; Luan Carvalho Staichack; Patrick Emami; Christopher Mauldin; and Pruthvi Manjunatha
Prepared by University of Florida Transportation Institute for Florida Department of Transportation