Welcome!
The Sustainable Mobility and Advanced Research in Transportation (SMART Lab) Lab focuses on leveraging the power of Big Data Analytics, Artificial Intelligence (AI), and the Internet of Things (IoT) to design and deploy cutting-edge technologies in transportation engineering. Our research aims to create efficient, safe, and sustainable transportation systems by integrating advanced technologies, such as connected and autonomous vehicles, traffic flow optimization, and infrastructure capacity planning. We specialize in handling and analyzing large-scale transportation datasets, developing machine learning techniques for predictive modeling, and exploring the potential of IoT devices to enhance infrastructure monitoring and traffic safety. Through interdisciplinary collaborations, active grant-seeking, and dissemination of our research findings in respected journals and conferences, we strive to make significant contributions to the field and secure funding for future projects.
Examples of our work
Pavement distress evaluation is essential for routine pavement maintenance. This process entails determining the type of pavement distress and assessing its severity via manual or computer vision techniques. Recent efforts have shown a steady increase in the performance of deep machine learning models for automating pavement distress analysis. These models are, however, data hungry; requiring training on large, diverse annotations to enable it to perform the same tasks on data from different sources. Existing datasets annotated for deep learning models lack diversity in their data sources, which usually results in problems of model transferability and over-fitting. The current study seeks to establish a new dataset that can be used to benchmark and compare new machine learning models developed for detecting and analyzing pavement distress.
The dataset was compiled using data from a variety of sources including Google Street View and pavement distress survey vans. To outline the distress boundary in the new dataset, polygons were used to annotate the images instead of bounding boxes. This enables the direct calculations of distress severity. A semi-automated annotation pipeline was developed to increase the size and variety of annotations. The study annotated over 13,000 distresses, categorizing them into seven distinct pavement types. Finally, six different deep U-Net based architectures were trained on the benchmark dataset to evaluate their performance. Based on validation intersection-over-union (IoU) score and predictive performance on test data, the resnest50 and resnet34 backbones were determined to be the most suitable U-Net backbones for distress detection using this dataset.
News
25 June, 2023 Congratulations on Joining University of Arizona as Assistant Research Professor.
17 April, 2023 Two papers accepted for presentation at CVPRw 2023.
1 January, 2023 Join Northwestern University as a Postdoctoral Student under the Supervision of Dr. Ulas Bagci.
17 December, 2022 Graduated with a PhD from the University of Missouri-Columbia.
11 November, 2022 Driver Maneuver Detection and Analysis using Time Series Segmentation and Classification was accepted for publication
12 October, 2022 Mobile Sensing for Multipurpose Applications in Transportation was accepted for publication.
20 June, 2022 Oral presentation at CVPR: A Region-Based Deep Learning Approach to Automated Retail Checkout
