The Classification of Skateboarding Tricks by Means of the Integration of Transfer Learning and Machine Learning Models

Generally, the assessment of skateboarding tricks executions is completed abstractly dependent on the judges’ understanding and experience. Hence, an objective and means for assessing skateboarding tricks, especially in the big competition are important. This research aims at classifying skateboarding flat ground tricks, namely Ollie, Kickflip, Shove-it, Nollie Front Shove and Frontside 180 through camera vision and pre-trained convolution neural network for feature extraction coupled with a conventional machine learning model. An amateur skateboarder (23 years of age ± 5.0 years’ experience) executed five tricks for each type of trick repeatedly on a skateboard from a camera with a distance of 1.26 m. From the images captured, the features were engineered and extracted through Transfer Learning, particularly VGG-16 and then classified by means of Logistic Regression (LR) and k-Nearest Neighbour (k-NN) models. The observation from the preliminary investigation demonstrated that through the proposed methodology, the LR and k-NN models attained a classification accuracy of 99.1% and 97.7%, on the test dataset, respectively. It could be shown that the proposed strategy can classify the skateboard tricks well and would, in the long run, support the judges in providing an increasingly objective-based judgment.