Vision-based classification of outdoor environments

Problem

The problem this project tries to solve is the lack of information on the surrounding areas found on the majority of real estate listings. That is, many details are given about the property itself and its features, but there is often a lack of information regarding the features and quality of the surroundings of the property.

Solution

We implemented a program that helps with the classification and prediction of natural surroundings and, mainly, urban areas, based on images.

How we built it

The code uses the Computer Vision models provided by Restb.ai, and images from the Street View, pulled from the Google Maps API.

Challenges we ran into

Some of the features that the model predicts aren't well-suited enough for exterior surroundings. The reason is that the endpoints used from the Restb.ai API are focused on working with real estate indoor properties.

Accomplishments that we're proud of

We implemented an innovative idea which has applications in many different areas, also outside of the real estate field. For example, it could also help identify possible improvements of urban areas. Another possible application is the optimization of transport routes (as an example, customers would be much happier to pass through a more natural or "green" sorrounding, than one lacking significant natural light).

What we learned

Among many other things, how to implement an idea in a really short time.

What's next for Vision-based classification of outdoor environments

This model would be greatly expanded by implementing an automatic way of getting street addresses.

We believe the next step of the project is to create a visualization for the information, for it to be displayed more intuitively and be used more easily to solve the different problems that inspired its creation.

Built With

• google-maps
• python
• restb-ai-api