Utilising computer vision to create an augmented reality training aid for basketball shooting : a thesis presented in partial fulfilment of the requirements for the degree of Master of Engineering at Massey University, Manawatū, New Zealand
Loading...
Files
Date
2022
DOI
Open Access Location
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The Author
Abstract
The use of Augmented Reality (AR) technology can improve an athlete's capabilities and help physical trainers add value to their teaching. Recently, the addition of technology in sports has come in many forms, from wearable sensors to monitoring the user's health and fitness and mapping their surroundings to virtual environments, which provide opportunities for practicing sport in remote locations, in isolation, and with additional feedback not available in real world practice. The goals of this research were to design and examine the feasibility of an immersive, fun and interactive experience through which the user can shoot a real basketball in an AR environment where the real hoop or target is replaced by a virtual target. The object tracking capabilities of HoloLens proved to be insufficient for ball tracking in this project due to its low capturing speed and limited accuracy at longer distances. As a result, the Intel RealSense D455 device was used with the HoloLens in an attempt to achieve better ball tracking and then effectively communicate the 3D position data to the HoloLens. The results section records the accuracy of tracking the static and dynamic centroid position of the ball in 3D space. The static results yielded accurate centroid values at 1m; however, beyond 2m, the noise affected the accuracy of the values. From the dynamic results, using two larger distinct regions of interest (ROIs) within 2m from the camera yielded the most accurate launch velocity predictions. UDP worked reliably to send this 3D data as strings to the HoloLens, which then transformed the 3D data using the ICP algorithm to calibrate the axes of the D455 and the HoloLens. The ICP algorithm had small inaccuracies when calibrating the two axes because it is commonly used to calibrate large data sets; however, all the mean error values between the transformed coordinates and known coordinates for all the axes fell within one standard deviation of the mean error. The final trajectory experiment was successful in spawning a ball at a 3D location and produced the most accurate results when hitting a target 0.8m to 1.2m away from the D455, however a final trajectory prediction analysis incorporating the launch velocity error measured in this work clearly demonstrated the technology is not currently suitable for use in long distance applications such as a typical basketball free throw. The current limitations of this tracking method and its efficiency as an AR training tool were explored and future recommendations were provided for better tracking accuracy in AR using the HoloLens.
Description
Figures are re-used with permission.