Quantifying drivers-overtaking-bicyclists with surrogate safety measures from a high-resolution 3D lidar

Rayane Moustafa
rayanem@umich.edu

University of Michigan-Dearborn

Gandhimathi Padmanaban
gmathi@umich.edu

University of Michigan-Dearborn

Fred Feng*
fredfeng@umich.edu

University of Michigan-Dearborn

January 14, 2026

Bicycling in the U.S.

  • Bicycling is underutilized with 0.5% of commuters riding a bicycle (vs. 69.2% commuters driving alone).1
  • 1,166 bicyclists were killed and 49,989 injured in 2023.2
  • Drivers overtaking bicyclists is one of the most dangerous encounters with rear-end collisions accounting for 40% bicyclist fatalities.3
  • Previous studies on overtaking bicyclists typically used lower resolution sensors (e.g., ultrasonic sensors4 or 2D lidar5), offering limited insights.

High resolution 3D lidar

  • Lidar (Light Detection and Ranging) is a laser-based sensor that captures 3D representations of its surroundings.
  • It enables detailed examination of road user behaviors and interactions (e.g., drivers, bicyclists) in real-world settings.

3D lidar

Visualization of point cloud captured by a 3D lidar

Study objectives

  • Collecting and sharing bicycle riding data in real-world settings using high resolution lidar and cameras.
  • Develop algorithms to automatically
    • detect and track overtaking vehicles based on the lidar point clouds captured on a bicycle,
    • compute quantitative surrogate safety measures for bicycling safety research.

The research bicycle equipped with a lidar (Ouster OS1 128-channel)

Demonstration of the point cloud collected from the lidar bicycle

Preliminary data collection

  • Two trips in Ann Arbor, MI
  • 4-lane arterial road
    • 35 mph speed limit
    • painted bike lane on each side
  • Bicycle riding behavior
    • stay in the middle of bike lane
    • riding speed: 12-15 mph
  • 306 vehicle overtaking events

Trip route (out and back)

Forward camera field of view

Point cloud and camera data collection demo

The overtaking vehicle detection and tracking algorithm

Sedan

SUV

Pickup truck

Cement truck

Examples of segmented vehicles with 3D bounding boxes

Surrogate Safety Measures (1/3)

Begin of passing

End of passing
  • Passing distance: the closest distance from the lidar to the car within the passing region
  • Passing duration: the duration of time that a vehicle spends in the bicycle’s passing region

Surrogate Safety Measures (2/3)

  • Bicycle speed: Based on frame-to-frame registration with
    • RANSAC (Random Sample Consensus6)
    • ICP (Iterative Closest Point7)

Surrogate Safety Measures (3/3)

  • Vehicle Speed
    • Relative vehicle speed estimated using numerical differentiation of vehicle centroid locations across frames
    \[ v(t)=\frac{x(t)-x(t-\Delta t)}{\Delta t} \]

\[ \text{vehicle speed} = \text{relative vehicle speed} + \text{bicycle speed} \]

Results

  • 306 overtaking events from two trips
  • 98 events had valid C3FT8 readings (ground truth).
  • Passing distance validation: MAE: 3.9 cm, MAPE: 2.5%

Camera view with C3FT reading (in blue)

Point cloud algorithm result


Validation of bicycle speed estimation based on GPS speed ground truth

Passing distance distribution

Passing duration distribution

Passing speed distribution

Passing speed vs distance

Discussions

  • Passing duration reveals risk exposure over time
    • Longer overtakes typically due to traffic congestions or large vehicles (e.g., commercial trucks)
  • The algorithms enable the detection of dangerous overtakes (i.e., close pass at high speed)
  • Overtaking behavior depends on a combination of factors including roadway design, vehicle characteristics, and traffic conditions.

Future work

  • Continued data collection effort covering more roadway types and traffic conditions.
  • Refining methods to improve vehicle detection accuracy under challenging conditions (e.g., tilted bicycle, roadway curvature, intersections)
  • Developing and analyzing additional overtaking safety measures
    • e.g., vehicle lateral movements, the timing selection to initiate an overtaking

Acknowledgements

This material is based upon work supported by the National Science Foundation under Award Number 2142757.

“CAREER: Improving Bicycling Safety by Developing a Research Framework for Studying Driver-Bicyclist Interactions”

References

1.
United States Census Bureau. Commuting characteristics by sex. Census.gov (2024).
2.
National Center for Statistics and Analysis. Bicyclists and Other Cyclists: 2023 Data. https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/813739.pdf (2025).
3.
League of American Bicyclists. Bicyclist Safety Must Be a Priority Findings from a Year of Fatality Tracking - and the Urgent Need for Better Data. http://bikeleague.org/sites/default/files/EBC_report_final.pdf (2014).
4.
Beck, B. et al. How much space do drivers provide when passing cyclists? Understanding the impact of motor vehicle and infrastructure characteristics on passing distance. Accident Analysis & Prevention 128, 253–260 (2019).
5.
Dozza, M., Schindler, R., Bianchi-Piccinini, G. & Karlsson, J. How do drivers overtake cyclists? Accident Analysis & Prevention 88, 29–36 (2016).
6.
RANSAC documentation. Readthedocs.io (2021).
7.
ICP Open3d documentation. Open3d.org (2017).
8.
C3FT. C3FT v3 – codaxus LLC. Codaxus.com (2017).

Resources