Authors:

Nishant Vasan

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

With the rise of Artificial Intelligence, there is a constant pursuit in sports to enable Computer Vision techniques to assist in training and umpiring. Table Tennis, being an extremely fast sport with ball speeds reaching up to 70mph, presents a significant challenge in tracking the ball's dynamics. Ball tracking could revolutionize training and umpiring. For instance, integrating the technology into a Virtual Reality setup could provide players with unique insights to accelerate the learning process. The objective is to build a system to estimate the 3D location of the ball. The data for the model was gathered from videos by two cameras. The ball is detected in 2D from the videos using a Blob Detection Algorithm and the hyperparameters are tuned to detect a white ball. The cameras are then calibrated and the projection matrices are estimated. Using Linear Triangulation, we then reconstruct its 3D position from 2D position. The results indicate that the system can estimate the 3D location of the ball and provide visualizations for its trajectory. Additionally, we implemented a marker based spin tracking algorithm as well. We use convolutional neural networks to locate the markers and use Bayesian Geometric Hashing to estimate the orientation of the ball. The relative orientation between consecutive frames helps us find the spin of the ball. Keywords: Table Tennis; 3D Position Triangulation; Object Detection

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Nishant Vasan