Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

89 Views

Abstract for presentation (Poster or Podium) with a Paper in the Conference Proceedings

AI in Transportation

A Deep Learning Approach for Multiple Weather Detection Based on In-Vehicle Dash Camera Video

Monday, June 17, 2024

1:30 PM - 2:45 PM ET

Room: Galleria

Primary Author(s)

Lizhe Li, PhD

Post-doc
University of Cincinnati
Cincinnati, Ohio, United States

Co-Author(s)

Mohamed Ahmed, PhD., P.E.

Professor and Director
University of Cincinnati
Cincinnati, OH, United States

Presenter(s)

Mohamed Ahmed, PhD., P.E.

Professor and Director
University of Cincinnati
Cincinnati, OH, United States

Submitter(s)

LL

Lizhe Li, Ph.D.

University of Cincinnati
Cincinnati, Ohio, United States

Abstract:

Weather conditions significantly influence traffic safety, with adverse conditions often leading to hazardous driving environments. This study presents the development of a weather detection model based on Convolutional Neural Networks (CNNs), specifically trained on in-vehicle dash camera videos. Using thousands of images sourced from the SHRP2 dataset, seven distinct weather categories were identified: clear, light snow, heavy snow, light rain, heavy rain, distant fog, and near fog. For enhanced training efficiency, images were strategically segmented. An innovative approach was employed where, on average, nine images were selected from every minute of the video, these images underwent processing using various pre-trained CNN architectures, such as 'AlexNet', 'GoogleNet', 'ResNet', among others. The predominant detection results then informed the overall weather condition for the given video segment. The model exhibited promising performance, underscoring its viability for real-time in-vehicle weather detection and consequent enhancement of road safety.

Learning Objectives:

Attendees can expect to learn the following from this session:

Upon completion, participant will be able to develop five pre-trained weather detection models using CNN networks for images from vehicle dash camera video.
Upon completion, participant will be able to compare the accuracy and performance of all pre-trained models with a full detailed list.
Upon completion, participant will be able to develop an image & video processing algorithm, utilize the SHRP2 dataset, and identify the weather using a pre-trained model.