Kinetic Motion Tracking of Basketball training motion data capture and analysis using Kalman Filter and DeepLabV3+

In the past few years, sports are emerging rapidly and developing all across the world and many technologies are embedded with respect to the advancements. Basketball is one of the most popular sports in the world. Analysis of sports plays a crucial role in enhancing the players performance as it records various parameters with respect to the sport. Basketball is also such a sport which on analyzing various factors such as acceleration, velocity and reaction time etc. can assist the players to know the areas on which they need to work. Even though there are existing methods such as Convolution Neural Network (CNN), Residual Neural Network (RNN) etc. still there are challenges in tracking the motion due to excessive noise and thus obtaining low accuracy. This paper proposes a method namely Median Filter Savitzky Golay Filter Kalman Filter and Deep LabV3+ (MSGKF-DeepLabv3+) which is integrated using numerous techniques on basketball training sports data i.e. Basketball Action Dataset (BAD). For Noise reduction at the pre-processing stage, Median Filter (MF)- Savitzky Golay Filter (SGF) are utilized and then for identifying the players, YOLOV8 is imposed. Additionally, for pose estimation - DeepLabV3+ model is used because it has ResNet50 and further for pose tracking, Kalman Filter is employed. Ultimately, MSGKF-DeepLabv3+ outperforms conventional methods such as CNN and RNN with an accuracy of 98.71%.