RouteVision

Inspiration

With the Super Bowl coming up this Sunday, we wanted to make a program that can help quarterbacks make better decisions. This application could be used at all levels of the sport, but could be especially useful for middleschool, highschool, and college students, who may not have access to professional level coaching to review film and improve their skills.

What it does

RouteVision analyzes NFL game footage and automatically identifies key moments in a passing play where a quarterback must make a decision.

At those moments, the system:

Detects and tracks all players directly from video
Differentiates offense and defense
Highlights each eligible receiver
Visualizes passing options with arrows and an open-window percentage
Provides a short, AI-generated explanation for why a receiver is a good or poor option

The result is an intuitive, explainable breakdown of quarterback reads based purely on vision and spatial context.

How we built it

We fine-tuned YOLOv8 on broadcast football imagery to reliably detect fast-moving, small, and motion-blurred players.
We used ByteTrack to maintain player identities across frames and stabilize tracking.
A computer vision pipeline computes geometry based features such as separation, closing speed, and throwing lane obstruction.
The system automatically selects important moments in a play by detecting peaks in receiver openness.
We use Backboard.io to orchestrate calls to Gemini, which converts these vision derived features into an interpretable open window percentage and a concise, coach style explanation.
A web interface allows users to upload clips, watch playback pause at key moments, and interactively explore passing options.

Challenges we ran into

Broadcast footage is adversarial for vision models: players are small, blurred, partially occluded, and mixed with crowd noise.
Tracking stability was difficult when detections briefly dropped during fast motion or camera pans.

Accomplishments that we're proud of

We are proud of our ability to persevere, debug, and problem solve when faced with errors
Completing a project from start to finish that we can see being useful to people as well as learning and using new interesting technologies

What we learned

We learned new technologies like YoloV8, ByteTrack, and many other data visualization and analysis tools
Developed our teamwork, communication, brainstorming, as well as many other integral skills for working in teams to successful build and deploy projects