<aside> 💡

My TG project aims to create an open-source real-time tennis line caller and analyzer that I can carry to the court. I’m designing and building a prototype that hopefully others can replicate and perhaps improve upon.

</aside>

Where I am today:

• I have a portable prototype that records video and can be controlled by an Android app. The above is an image captured by the TG prototype. I’ve created, trained, evaluated a convolutional neural net that should be able to accurately detect tennis balls in the video frames. The net currently just runs offline, but should eventually be able to run in real time on my prototype.

Progress Updates

Hardware Ordered and put together (10/9/25 - 11/13)

Software Setup — getting to 2 working cameras (10/13- 11/6)

Networking and an Android App

Video Recording and Encoding

4K vs HD

Tennis Genius Placement

Finding a tennis ball in the video (Dec ‘25)

Labelling data and an initial CNN classifier (Dec ‘25)

Quick YOLO Experiments (Jan ‘26)

A trainable detector: the FCN (Jan ‘26)

Really Labelling my videos (for tennis balls) (Jan ‘26)

Tweaking the FCN (Feb ‘26)

Serve Detection (Feb ‘26)

The return of HD! (Feb ‘26)

What’s next

What’s missing