3D digital twin modeling of arterial traffic flow using trajectory data from roadside LiDAR for C-V2X applications

Cellular Vehicle-to-Everything (C-V2X) is the prevailing technology of Connected Vehicle (CV) technologies that uses cellular 4G or 5G network to deliver CV Society of Automotive Engineers (SAE) J2735 and J3224 messages to enable a wide range of traffic safety, mobility, and environmental applications as recommended by the U.S. Department of Transportation. Recently, with the rapid development roadside high-resolution sensing technologies, such as LiDAR (Light Detection and Ranging) and deep-learning-based computer vision sensors, the deployment of C-V2X applications no longer relying on trajectories reported by on-board units (OBUs) in vehicles but based on the complete trajectory data from roadside sensors allowing the "Day 1" application deployment schema where CAV applications are be turned on instantly for any users rather than waiting for slow increase of OBU penetration into the traffic. In such a Day-1 deployment scheme, High-resolution sensors such as roadside LiDAR can detect the trajectories of vehicles, pedestrians, and other road users, and report high-resolution trajectory information on behalf of all detected objects to passing vehicles, including those not yet equipped with communication systems, in accordance with sensor messaging standards through physical and virtual roadside units (RSUs).However, high-resolution data does not equate to high-quality data. The current vehicle trajectories from roadside LiDAR have three main issues: First, due to fixed perception blind spots limited by the field-of-view constraints when migrating onboard LiDAR sensors to roadside applications, the trajectory loses tracking at specific locations such as the area underneath the LiDAR; Second, occlusion caused by static infrastructure and dynamic vehicle blockage generates trajectory fragments at random positions; Third, deep-learning object detection and tracking instability further degrades trajectory continuity. Additionally, existing trajectory association and completion methods perform poorly under oversaturated or shockwave conditions on urban arterial roads; deep learning approaches are predominantly self-regression-based and do not account for inter-trajectory associations; and many methods lack the computational efficiency required for real-time applications. These issues fundamentally limit the confidence in data quality necessary for mission-critical C-V2X safety and mobility applications that demand high-resolution trajectory data. This dissertation focuses on reconstructing the trajectories of road users in urban corridors with large-scale multi-sensor deployment to support the 3D digital twin modeling of arterial traffic flow. The research first generates a digital twin of the infrastructure and roadway and develops a waypoint-based centerline map as the global coordinate reference for trajectory completion. A vehicle trajectory completion method is then developed based on shockwave boundary detection from roadside LiDAR, which classifies trajectories by their spatial completeness relative to the sensor's detection range and employs a bidirectional search to determine forward and backward trajectory pairs. Shockwave boundaries are detected through trajectory curvature analysis, coordinate transformation, histogram-based clustering, and linear regression, enabling trajectory interpolation that respects the underlying traffic dynamics, including queuing and discharge phases. For cross-sensor trajectory reconstruction, the method extends to associate trajectory segments from sensors across multiple sites using the unified traffic lane waypoint map system. The proposed methods are evaluated using trajectories extracted from industrial edge computing from the DataCity Smart Mobility Testing Ground project, covering intersections equipped with high-resolution roadside LiDAR sensors on the Route 27 corridor in downtown New Brunswick. Evaluation results demonstrate that the proposed method reduces rapid ID switching across different time-of-day periods, outperforms the reference minimum-cost circulation method in both trajectory stitching count and interpolation accuracy under congested conditions, and produces reconstructed trajectories that align well with camera-based space-time map validation. Comprehensive calibration and sensitivity analyses on spatial and temporal gap thresholds, point density effects, and shockwave parameters have been conducted to assess the robustness and transferability of the proposed model. The reconstructed trajectories are integrated into a 3D traffic digital twin model, enabling real-world high-fidelity traffic visualization, connected automated vehicle application modeling, and macroscopic operational analytics along the urban corridor.