Action Recognition of Skateboarding Tricks – Ollie and Kickflip Using Neural Network

This research project aimed to develop a computer-based system using deep learning techniques to accurately detect and recognize skateboarding tricks, with a focus on ollies and kickflips. A deep learning architecture that combined RCNN, Mobile Net with Bi-directional LSTM, and CNN was proposed and implemented on a dataset of 222 skateboarding trick videos categorized into two subdirectories - Ollie and Kickflip. The proposed models were trained to precisely identify the different skateboarding motions, and the results of the experiment showed how well the recommended deep learning model classified ollies and kickflips. The trial results revealed that the algorithms were highly effective in guiding viewers through the process of scoring skateboarding films. Based on their accuracy, precision, recall, F1-Score, and AUC, the three deep learning models CNN, CRNN, and Mobile Net with Bidirectional LSTM were assessed. The results showed that CRNN had the highest accuracy and AUC of 79% and 86 respectively, while Mobile Net with Bidirectional LSTM had the highest recall and F1-Score.