Modeling lane-selection and lane-change frequency on congested urban freeways: empirical evidence from Chicago’s I-90/I-94 freeway

Yousefi, Mahdi

Permalink

https://hdl.handle.net/2142/130231 Copy

Description

Title

Modeling lane-selection and lane-change frequency on congested urban freeways: empirical evidence from Chicago’s I-90/I-94 freeway

Author(s)

Yousefi, Mahdi

Issue Date

2025-07-25

Director of Research (if dissertation) or Advisor (if thesis)

Talebpour, Alireza

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Lane-Changing Behavior
• Vehicle Trajectory Analysis
• Multinomial Logit Modeling

Abstract

Lane-changing behavior plays a significant role in traffic movement. It significantly impacts how well traffic flows, the safety of the roads, and the overall functioning of the system. This is a comprehensive study of lane-change behavior on a major urban freeway segment, focusing on vehicle trajectory data collected via aerial videography. The research utilized a 4-kilometer segment of the I-90/I-94 freeway in Chicago, Illinois, with data gathered during the afternoon using a helicopter-mounted camera at an altitude of 200 meters. The study aimed to uncover how traffic conditions, vehicle types, and segment-specific factors influence drivers’ lane choices. The data collection methodology involved tracking target vehicles over multiple trips using high-resolution video and extracting vehicle trajectories through a six-step process: image preprocessing, object detection, object tracking, image stabilization, trajectory construction, and data cleaning. The combination of RetinaNet and YOLOv5 models enabled the accurate detection and tracking of vehicles in both moving aerial and infrastructure-based video footage. Vehicle positions were stabilized using reference satellite imagery, feature-matching techniques (SIFT and FLANN), and homography transformations. The final trajectories were refined through Kalman filtering, manual ID stitching, and linear interpolation to ensure continuity across segments. The six-step extraction process provides detailed and reliable trajectory datasets, which are essential for analysis. The study explored the relationship between average speed and lane-change frequency using polynomial regression. Lane changes were most frequent at moderate average speeds around 14 kilometers per hour and decreased at both lower and higher average speed ranges. A third-degree polynomial model fit the data well, indicating that drivers change lanes most often during moderate congestion, as switching lanes helps them gain more speed. Similarly, a density-based analysis revealed that lane-change frequency increases sharply up to a critical density of 54 vehicles per kilometer per lane and then decreases as traffic becomes more congested. This pattern was also well captured by a third-degree polynomial, indicating that drivers change lanes more frequently at moderate traffic levels but do so less often in heavy traffic due to the limited availability of gaps. A joint analysis of density and average speed using contour maps revealed that lane changes peaked at 75 vehicles per kilometer per lane and 17.5 kilometers per hour. The analysis showed that lane changes vary by position along the freeway. Within the first 900 meters, most drivers shifted to the right to prepare for the I-94 exit. Beyond 900 meters, a multinomial logit lane-selection model is used, with density and average speed as predictors. This model estimated the probability of choosing each of six lanes and generated probability surfaces across the traffic state space. It confirmed that, accepted space gaps under 20 meters, while large vehicles preferred gaps of 20-50 meters and changed lanes more cautiously. These patterns reflect operational constraints such as braking distance and visibility. Time gap analysis showed that over 70% of all lane changes occurred when the available time gap was less than five seconds, with small vehicles making the majority of these quick maneuvers. The study also developed a multinomial logit lane-selection model using density and average speed as predictors. This model estimated the probability of choosing each of six lanes and generated probability surfaces across the traffic state space. It confirmed that in free-flow traffic, drivers prefer the outer lanes. The model's closed-form structure supports real-time traffic simulation and control. Key findings support the idea that lane-changing is a context-dependent behavior driven by traffic conditions, vehicle type, and upcoming route decisions. The modeling approach provides a practical tool for predicting lane-use patterns in real-time traffic management systems.

Graduation Semester

2025-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/130231

Copyright and License Information

Copyright 2025 Mahdi Yousefi

Owning Collections