Research and Development Ball Tracking

Deep learning

A demonstration of the type of results achievable with deep learning, here with some tennis shot prediction. The “picture-in-picture” is the raw computer vision output, with predictions in the main window as blue traces. As you can see, some false positives and negatives. Although, some of the misses are due to the computer vision software rather than the prediction filter, which in turn is down to the poor video quality.

Non-machine learning approach

Embedding intuition from knowledge of the problem domain as an early alternative to machine learning.