Street Smarts

Inspiration

UPS instructs its drivers to avoid left turns at all costs in order to reduce the amount of stopping needed, which saves lots of fuel and money for the company and reduces the total carbon emissions from their delivery trucks. We thought that AI could reduce the time vehicles need to spend stopped at a red light, and therefore drastically reduce the total auto emissions on a national or global scale. Another main inspiration for this project was the fact that it requires no outside/extra effort from the driver. Instead, it actually is more convenient for the drivers to drive in the Street Smarts system, as they spend less time at red lights.

What it does

Street Smarts uses AI with already existing red light cameras to detect the number of vehicles going to or from an intersection. We can obtain data via Google Maps to connect intersections to each other and determine how many vehicles are heading towards a new intersection. Then, using a deep neural network with tensor flow, the individual intersection can decide the most efficient traffic pattern based on the predicted incoming vehicles and the detected vehicles from the red light cameras.

How we built it

We first used HTML and JavaScript to create a simple example intersection with cars, and measured the average efficiency using a timer system (similar to the real world's current system). Then we wrote a neural network with tensor flow in Javascript (originally in Python) to control the traffic light and optimize traffic flow (cars through the intersection per second).

Challenges we ran into

One of the unexpected challenges was making the "cars" act and behave as a real driver would (maintaining a following distance, stopping at a red light, accelerating after stopping, etc.). However, the main challenge we ran into was making the neural network controlling the traffic light better at predicting traffic patterns and optimizing the traffic flow.

Accomplishments that we're proud of

We are very proud of how the neural network can make traffic more efficient, as we know with more work and data, it could ideally be implemented into the real world and greatly reduce auto emissions.

What we learned

We learned how to use Tensorflow with Javascript. It was super cool learning how to use machine learning in a browser! We also learned a ton about swarm intelligence while working on making the car simulation. And we definitely learned a ton about HTML and Javascript.

What's next for Street Smarts

We would like to get enough camera feeds to be able to integrate our openCV code with the visualization. We would also like to do more testing on larger networks of roads and different types of AI.