1. Home
  2. RIP
  3. View Record
https://trb-annual-meeting.nationalacademies.org/registration

A Deep Learning Approach for Detecting Built Environment in Transit-Oriented Developments

Transit-Oriented Developments (TODs) are a pivotal strategy for fostering sustainable urban growth amidst escalating urbanization and population spikes. These developments are strategically designed to combat urban sprawl, significantly reduce the reliance on automobiles, cut greenhouse gas emissions, and create more livable, community-focused urban areas, and help address the housing needs. The core philosophy behind TODs lies in integrating residential, business, and leisure spaces with public transportation systems, thereby promoting a lifestyle that prioritizes walking, cycling, and the use of public transit over automobile use. The effectiveness of TODs in achieving these goals hinges on the built environment’s configuration, which influences sustainable transportation adoption, boosts economic vitality, ensures accessibility, fosters social equity, and champions environmental stewardship. However, the acceleration of TODs brings forth a pressing challenge: How to accurately and efficiently measure the built environment in TODs across a large geographical region. To tackle this challenge, this project proposes a novel deep learning framework to automate the detection and categorization of essential built environment elements within TODs. By leveraging the well-established 5D framework—density, diversity, design, destination, and distance to transit—this framework seeks to comprehensively identify and map the components that constitute an effective TOD. To accomplish this, the research project will develop a sophisticated deep learning architecture capable of assimilating multi-sourced datasets with various modalities (e.g., imagery, text, tabular, GIS data). These will include high-resolution aerial imagery to capture urban layouts and green spaces; Google Street-View imagery for a pedestrian-level perspective of the urban landscape; parcel and OpenStreetMap data for detailed insights into land use, infrastructure, and building outlines; General Transit Feed Specification (GTFS) data for a comprehensive overview of public transit networks; and Census data to incorporate demographic insights. This wealth of information will be processed using a blend of cutting-edge machine learning techniques, including but not limited to, pretrained Convolutional Neural Networks (CNNs) and attention-based Transformer models for decoding unstructured data (such as images) and Graph Neural Networks (GNNs) for processing structured data analysis (such as GIS and tabular data). The proposed framework aims to create a nuanced and comprehensive understanding of TODs’ built environments, facilitating the detection of key features like buildings, pedestrian crossings, transit lanes, green spaces, and more. A database created by the PIs will serve as the primary case studies for this research, providing a diverse range of urban contexts for evaluation and validation of the proposed framework. The project outlines several major tasks: data acquisition, preprocessing, development and validation of the deep learning model, applying the validated model to additional TOD locations, hosting educational workshops, and compiling findings into a final report. The proposed approach will be developed, evaluated, and validated by using randomly selected TOD locations across Florida. This research endeavors to equip urban planners, transit authorities, and policymakers with an advanced tool for automatically identifying critical elements of TODs’ built environment, thereby facilitating smarter, more sustainable decision-making nationwide. Outputs will include 1) Publications & conference contributions 2) Database: An open-source database, including various sources of downloadable data for TOD analyses, to disseminate the research outcomes and encourage future research. This database can be used to generate interactive maps, visualizations of the built environment's impact on equity and accessibility within TODs, & will serve as a resource for urban planners, policymakers, researchers, and the public. 3) Methodologies & Technologies: The project will introduce innovative deep learning architectures that can be directly used to automatically detect and assess the built environment elements within TOD. Code repositories and fine-tuned models’ weights will be open sourced to keep transparency and allow for reproducibility for future research. 4) Partnerships: Establish and enhance partnerships with stakeholders from local government, non-profits focused on urban equity and environmental justice, & tech companies specializing in geospatial data and AI.

Language

  • English

Project

  • Status: Active
  • Funding: $300000
  • Contract Numbers:

    69A3552348337

  • Sponsor Organizations:

    Office of the Assistant Secretary for Research and Technology

    University Transportation Centers Program
    Department of Transportation
    Washington, DC  United States  20590
  • Managing Organizations:

    Center for Equitable Transit-Oriented Communities (CETOC)

    University of New Orleans
    New Orleans, LA  United States 
  • Project Managers:

    Kline, Robin

    Danton, Bob

  • Performing Organizations:

    University of Florida

    207 Grinter Hall
    PO Box 115500
    Gainesville, Florida  United States  32611

    Florida Atlantic University, Boca Raton

    Boca Raton, FL  United States  33431
  • Principal Investigators:

    Zhao, Xilei

    Renne, John

  • Start Date: 20241001
  • Expected Completion Date: 20250930
  • Actual Completion Date: 0
  • USDOT Program: University Transportation Centers Program

Subject/Index Terms

Filing Info

  • Accession Number: 01928834
  • Record Type: Research project
  • Source Agency: Center for Equitable Transit-Oriented Communities (CETOC)
  • Contract Numbers: 69A3552348337
  • Files: UTC, RIP
  • Created Date: Aug 26 2024 2:51PM