Yanlin Zhang

Can the Waymo Open Motion Dataset Support Realistic Behavioral Modeling? A Validation Study with Naturalistic Trajectories

Mon, 08 Sep 2025 00:00:00 +0000

An Investigation of Discretionary Lane-changing Decisions: Insights From the Third Generation SIMulation (TGSIM) Dataset

Mon, 07 Apr 2025 00:00:00 +0000

Can the Waymo Open Motion Dataset Support Realistic Behavioral Modeling? A Validation Study with Naturalistic Trajectories

Sat, 08 Feb 2025 00:00:00 +0000

Introduction to the Third Generation Simulation Dataset: Data Collection and Trajectory Extraction

Wed, 08 Jan 2025 00:00:00 +0000

Advancing Self-Enforcing Streets Phase 1: The Relationship between Roadway Environment and Crash Severity

Sun, 08 Dec 2024 00:00:00 +0000

Characterizing Human--Automated Vehicle Interactions: An Investigation into Car-Following Behavior

Sat, 25 May 2024 00:00:00 +0000

The Third Generation SIMulation Data (TGSIM)

Sun, 31 Dec 2023 00:00:00 +0000

Project Overview

The Third Generation SIMulation Data (TGSIM) project is a three-year research effort funded by the U.S. Department of Transportation (USDOT) to develop a comprehensive understanding of the impacts of automated driving systems on human behavior. The project is a led by Dr. Alireza Talebpour (UIUC), Dr. Hani Mahmassani (Northwestern) and Dr. Samer Hamdar (GWU).

The project aims to collect and process accurate trajectory datasets capable of characterizing human-automated vehicle interactions under a diverse set of scenarios in diverse highway and city environments. The datasets will be used to develop and validate traffic flow models that account for the impacts of automated driving systems on traffic flow. At the forefront of this project, I served as the leading graduate research assistant, playing a crucial role in developing a comprehensive data extraction pipeline that stands as a testament to our innovative approach to understanding the future of mobility.

Data Collecting Methodologies

Multiple methods were utilized to ensure comprehensive coverage of our study areas:

Fixed location aerial videography: a helicopter hovers over a segment of interest.
Moving aerial videography: a helicopter tracks the automated vehicles as they navigate a longer roadway segment.
Infrastructure-based videography: multiple overlapping cameras located on overpasses and buildings creating a comprehensive image of the study area.

Data Processing Pipeline

A sample of the extracted trajectories.

Extracting multiple vehicle trajectories from video data raises a set of methodological and practical challenges that vary across the three data collection approaches. The pipeline developed for this project includes the following steps:

Preprocessing:
- Raw Image Extraction: Converting the vedio into a sequence of frames separated at 30 fps over time.
- Reference Image Gereration: Developing a consistent methodology to convert each image into a fixed coordinate system (i.e., reference image).
Object Detection: Leveraging RetinaNet and YOLOv5 for precise vehicle identification within each frame.
Object Tracking: Performing a centroid tracking algorithm to trace vehicle trajectories through consecutive frames.
Image Stabilization
Trajectory construction

Impact and Contribution

This data collection and processing approach enabled us to extract more than 11,000 vehicle-kilometers of high-fidelity trajectory data from 14 hours of video footage. My contributions not only facilitated the development of a data processing pipeline that is of precision but also paved the way for a deeper understanding of automated driving systems’ impact on traffic flow and human behaviors.

The datasets, poised to be released to the research community by June 2024 through the USDOT data portal.

Automated Vehicles for All (AVA)

Thu, 16 Nov 2023 00:00:00 +0000

Project Overview

The Automated Vehicles for All (AVA) project is a four-year research effort funded by the U.S. Department of Transportation (USDOT). This project is dedicated to forging a systematic approach to achieve a safe integration of automated driving systems into the existing transportation infrastructure, especially on rural roads and multimodal driving enviorments. The project is a led by Dr. Reza Langari (TAMU), Dr. Alireza Talebpour (UIUC), Dr.Francis Assadian (UC Davis) and Dr. Samer Hamdar (GWU).

Aiming to enhance the safety and efficiency of automated driving systems in rural and multimodal driving environments, this project endeavors to not only evaluate these systems, but also generate a comprehensive dataset for automated vehicle safety analysis and rulemaking. Within this project, I served as a graduate research assistant, contributing across multiple teams including perception, data analysis, and safety testing.

My contribution

Real-time LiDAR point cloud segmentation

Real-time road segmentation using LiDAR data (code): I undertook the evaluation of various deep learning models, ultimately integrating the SphereFormer model with ROS to facilitate real-time road segmentation utilizing LiDAR data. This application was rigorously tested in a live environment with an automated vehicle, showcasing the potential for enhanced navigational safety.
Comprehensive Data Extraction Pipeline (code): I designed and implemented a comprehensive data extraction pipeline that invinvorporates camera-based detection, 2D-to-3D data fusion, and object tracking to generate a detailed dataset for the analysis of automated vehicle safety and the formulation of regulatory standards.
Implementation of control algorithm and Satety Testing: I conducted safety testing on the control and acturation systems of automated vehicles for rural environments in Rantoul, IL.

Through these contributions, I aimed to push the boundaries of what’s possible with automated driving technologies, ensuring they can be safely and effectively integrated into our transportation systems, particularly in environments that present unique challenges with less infrustructure supports.