Skip to main content
ICTD 2025 Main banner
Final Program


Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

ICTD 2025

Data Sensing & Analytics

POSTER RECEPTION (View list of posters here)

Adaptive Cruise-Control Detection from Naturalistic Driving Data using Sliding-Window CNN

Tuesday, June 10, 2025
4:45 PM – 6:00 PM MT

    Co-Author(s)

  • RO

    Richard Omenyo, n/a

    PhD Student
    The University of Alabama in Huntsville
    Huntsville, Alabama, United States

    • Presenter(s)

  • RO

    Richard Omenyo, n/a

    PhD Student
    The University of Alabama in Huntsville
    Huntsville, Alabama, United States

    • Primary Author(s)

  • RB

    Rahul Bhadani, PhD

    Assistant Professor
    The University of Alabama in Huntsville
    Huntsville, Alabama, United States

    • Co-Author(s)

  • SS

    Sajal Das Shovon, n/a

    PhD Student
    The University of Alabama in Huntsville
    Huntsville, Alabama, United States

    • Submitter(s)

  • RB

    Rahul Bhadani, PhD

    Assistant Professor
    The University of Alabama in Huntsville
    Huntsville, Alabama, United States

Abstract: In this work, we present a novel approach to predicting the driving mode of a vehicle from an external observer's point of view using machine learning models trained on naturalistic driving data obtained from thousands of kilometers of human driving. We develop a sliding window technique to generate image-like data from a multivariate timeseries driving dataset to segment it into fixed-size input windows. Such an approach enables use to deep-learning techniques design for image classification to our multivariate timeseries data. Using this windowing strategy, temporal relationships and patterns hidden in the data can be uncovered, allowing the model to learn and accurately forecast cruise control status. A CNN architecture, designed for temporal convolution, processes the segmented data. The CNN model is trained to predict the adaptive cruise control status of the vehicle. The trained model is used for making inferences in real-time by emulating live-streaming data using a Robot Operating System (ROS) package that shows that our model can predict the driving in real-time. Such detection is necessary for designing an improved longitudinal autonomous control for mixed-autonomy traffic and human-in-the-loop detection. By advancing autonomous driving technologies, this research contributes significantly to improved road safety, reduced traffic congestion and enhanced mobility.

Learning Objectives:

Attendees can expect to learn the following from this session:
  • Upon completion, the participant will be able to learn how to use deep-learning-based techniques for multivariate time series segmentation and classification.
  • Upon completion, participants will be able to understand the statistical difference between various driving modes.
  • Upon completion, the participant will be able to learn more about mixed-autonomy traffic.