ProFastBowl: Machine Learning Powered Right-Arm Fast Bowling Technique Enhancement with Pose Estimation

Abstract

The analysis of cricket right-arm fast bowling techniques can be characterized as fundamental in improving performance and preventing injuries. But still the advanced systems employing the latest technologies and thus capable of validly assessing fast bowling biomechanics are lacking. The current methodologies typically involve manual or basic video-based systems, neither of which offers the profound insight needed for both player improvement and injury mitigation. The aim of this research is to create an application which gives real time analysis and detect defects of right arm fast bowling. A hybrid machine learning technique was utilized to accurately detect the body joints, drawing on the OpenPose and PoseNet algorithms with the help of neural networks, which have the capability for pattern recognition. Right-arm fast bowlers' video footage was preprocessed with feature scaling, outlier detection, treatment of missing values, etc. The key body movements of fast bowling were used for system training with a view to recognize them correctly. This will also make deeper biomechanical analysis possible. Previous work had limitations such as focus was only on leg spin bowling, few battings strokes analysis, accuracy concerns, real time feedback providing, etc. The system can process and analyze video frames in real time, therefore providing immediate feedback on the performance of a bowler, while it is also capable of analyzing up to 500 frames per second.