Our Approach

Initially, we were going to make a pair of smart glasses that identifies players and team pocessions and displays relevant live and historical statistics to the glasses. We then pivoted to making a CV model that canidentify teams and pocessions and a web app surrounding this. The web app allows users to upload videos and annotate the videos and display relevant live and historical statistics. The stats are from the SynergySports API and a local database we created. We also created a wrapper for the SynergySports API to demonstrate the potential of the API.

Work Log for Each Sprint

Sprint 1

The goal for sprint 1 was to get a basic computer vision model up and running. We wanted to be able to detect players in a video and track their movement. Also, we began collecting historical data and setting up a database to store it.

Embedded Database for Historical Stats
- Created a SQLite database to store historical stats
Basic CV Functionality
- Simple possession identification

Sprint 2

In sprint 2, we found out we were not getting any glasses. So, we pivoted to a web app that would allow users to upload videos and get the same functionality as if they were wearing the glasses. We also wanted to improve the accuracy of our computer vision model.

Created a custom dataset to train our model
- Annotated over 750 pictures
Used our new dataset to train a more accurate model
- Tried various sizes to optimize the speed-to-accuracy ratio
Converted the computer vision service to process a single frame at a time
- More accurately simulates a live feed
Developed a web app due to lack of hardware
- Demonstrates the computer vision functionality

Sprint 3

For Sprint 3, we focused on improving the accuracy of our model and adding more functionality to the web app. We also created a wrapper for the Synergy API to demonstrate the potential of the API. Addionally, we focused on documenting and setting up the project to be easily handed off to future teams.

Improved color detection
- Used to assign players to teams
Added more functionality to the front end of the web app
- Team name and color assignment
Created a wrapper for the Synergy API
- Demonstrates the potential of the API
New algorithm for assigning possession
- Scales boundary boxes based on depth
Post possession smoothing
- Corrects short bursts of misdetections