Neural Network

One of the goals of the Juggling Data Set is to use computer vision to determine which pattern is being juggled in a video sequence.

First the features, which are the juggling balls, have to be tracked. Using those tracked locations, images that show the feature trails are generated.

These images show the feature trails generated from three juggling patterns; left - cascade, middle - columns, right - two balls in the right hand.

The feature tracks were used as the data for the neural network. The model was generated using Keras and Tensorflow. An excellent tutorial can be found at Learnopencv.com.