Looking for Hot Signals in Your Area: Cell Tower Detection

❤️ Rating the best (and worse) cellphone towers ❤️Tired looking at droneshots to see if your cellphone tower is up to code? Fear no more! We rate cell phone towers just for you!

Comment

Inspiration

When I first heard about Verizon's prompt, it was daunting. But as I searched more into the topic and thought about how I would go about this idea of cell tower detection, I thought about a rating system. Of course, giving it plainly would be boring. So, the idea snowballed into the rating system being presented as "dating/matching" system.

How I Built It

Tech stack: Python, Flask, Gemini, Verizon's Computer Vision Model (YOLOv8), HTML/CSS/JS Given drone imagery, it uses computer vision to detect the types of antennas. This data and original imagery are sent to Gemini (Flash 2.0) to detect further damages and compile a rating from 1 to 5. A profile on the tower is then created and listed is all the anomalies it ran into.

Challenges I Ran Into && What I Learned

This is my first time using AI, Computer Vision, and more than basic python in my program before, so it was all very new to me. So, using and learning these new technologies was very daunting at first as well. However, I was able to create something which I'm very proud to show and able to say that I've learned how to work with these new technologies.

What's next for Looking for Hot Signals in Your Area: Cell Tower Detection

There are still many things that I could improve and complete on. For example, I believe learning the geolocation of the cell tower may improve the urgency or rating of a cell tower's maintenance. So, while this data isn't provided, it is something I would love to work on more as if more people use that cell tower, the higher the rating should be. I would also like to use or train a better model then Gemini to detect things. My idea is to use 3 different Yolov8 models to detect three things: types of antennae, odd objects, city/location of cell tower. This data can then be fed to another model or be given as straight data to be able to make better predictions for maintenance.

Built With

Share this project:

Updates