Research
Deep Learning Image Classification with Google Street Views (GSV)
Project Goal: Collect and analyze street-level images from downtown areas of San Diego and San Francisco to identify homeless activity using Google Street View (GSV) for informed policy decisions.
Sample Image from the data collected ( Location - 17th street, Downtown, San Diego, CA)
Sample metadata extracted using GSV API
Completed Tasks:
Set up Google Cloud account and obtained GSV API key.
Identified and extracted images from specific downtown coordinates.
Developed a structured metadata system for image organization.
Recorded timestamp, location coordinates, and bearing for each image, then analyze these datasets to assess changes in specific areas over previous years as part of our effort to address homelessness.
Challenges:
Managing a large volume of data efficiently.
Automating image extraction for multiple years and months.
Reducing runtime complexity for quicker data collection.
Next Steps:
Preprocess the collected data.
Collaborate with experts to annotate images with homeless activity indicators.
Curate a labeled dataset for future deep learning analysis.
Planned Algorithms: Employ deep learning algorithms like CNNs for automated homeless activity detection.
Impact: This project aims to contribute data-driven insights for addressing homelessness in urban areas.
Comparison of Images extracted between current and the previous yearÂ
(GIS coordinates used: 32.705251, -117.148961)
Dated - Oct 2022
Dated - Feb 2020