Progress Report

Progress Report
Author

Morenike Atejioye

Published

June 12, 2023

Housing Project Progress Report (Year 1)

The project’s initial phase involved establishing objectives that were aimed to be achieved. These include;

  1. Using data to find good study cities;

  2. Literature review on using AI in housing and planning.

  3. Demonstrate the potential of AI to address housing questions;

  4. Focus on gutter-related issues.

Can we use street view images, or photos taken by the DSPG Team?

Can we use the photos on the assessor database/website?

  1. What county data are available to assess housing conditions?

What other data might we use to assess housing conditions?

  1. Identify where local, state and/or federal funds might be best applied.

  2. Tools that will be useful in Year 2 and 3.

Our Progress

  1. Literature Review

We conducted research on the methodologies employed by cities and counties in evaluating housing conditions. Our research encompassed examining Ames city, Des Moines city, the U.S. Department of Housing and Urban Development, Orange County, and Detroit. Additionally, we compiled research on how cities are using AI.

  1. Image Gathering / Data Collection

  • Zillow

  • Realtors.com

  • County Assessor Pages

Vanguard: Independence

Beacon Schneider: Slater, New Hampton, and Grundy Center

We have obtained house images from the mentioned websites. Our focus is on four communities: Independence, Slater, New Hampton, and Grundy Center. However, during the process of web scraping to gather images, we encountered an issue with blurred houses. Further investigation revealed that certain homeowners pay Google to have their residences intentionally blurred on Google Street View.

To ensure convenient access to the images collected from various sources, we devised a standardized naming convention: Source, City, Address. For example,

- Zillow

- New Hampton, Iowa

- 311 W Main St

Result: Z_H_311 W MAIN ST_

Z - Zillow

G - Google

V - Vanguard

B - Beacon

S - Slater

H - New Hampton

D - Independence

G - Grundy Center

During the data collection process, we encountered “address” inconsistencies in Independence. Some houses had multiple house numbers listed, such as 100/101, while others had addresses like 100 1/2. Fortunately, the function used to retrieve Google images managed to run successfully, although it reported 50 errors related to these address variations, including both the dual house numbers and the presence of half signs. It is worth noting that even if one of the address numbers is removed or the 1/2 is eliminated (essentially removing the “/” sign), the image URL still leads to a valid Google image. In order to address this issue, it is possible to revisit the URLs and modify them to obtain images corresponding to these specific addresses, if necessary.

  1. Build, Train, and Test AI Models

  • Vegetation Model

  • Siding Model

  • Gutter Model

In the initial stages of the project, we developed a vegetation model to gain insights before completing the housing data collection. This model utilizes artificial intelligence to analyze images and determine the presence or absence of vegetation in front of houses, providing a binary output of 0 (no vegetation) or 1 (vegetation detected).

Moving forward, our objective is to construct additional AI models. The first model aims to identify whether a house is present in a given picture, assess if the image is obstructed by trees or cars, and determine its visibility. Additionally, this model will distinguish between images containing a single house or multiple houses.

We also plan to build models to address specific aspects such as vegetation, siding, and gutters. The vegetation model, which was developed earlier in the project, will be integrated to classify the level of vegetation as overgrown, maintained, or nonexistent. The siding model will assess whether the panels are broken or missing, if the paint has chipped off or faded, or if the siding is in good condition. Similarly, the gutter model will identify if the gutter is missing, damaged, or in good condition.

  1. Create Database of Housing Information

  • Zillow

  • Realtors.com

  • County Assessor Pages

Vanguard: Independence

Beacon Schneider: Slater, New Hampton, and Grundy Center

Currently, we have not established a comprehensive database to store all the data collected throughout the project. However, creating such a database would prove beneficial for the second year of the project. This centralized repository would enable efficient organization, management, and retrieval of the accumulated data, facilitating seamless access and analysis for future endeavors.

To summarize the task at-hand,

  1. Identify study communities (places with population between 500 - 10,000, stable or increasing population, have a school, etc).

  2. Create a database of housing information.

  3. Sort images.

  4. Build AI models.

  5. Compile all information in a detailed report.