Deepracket: AI powered player performance evaluation in racket sports

Abstract

The project, “Qualitative Racket Player Analysis” is to analyse the player performance in a professional racket sport match using advanced video analysis techniques. It is predictive of the most common angles to opposite floor locations. A thorough analysis like court line detection using image processing to determine accurate boundaries of courts, player recognition and motion tracking during matches via particle filter algorithms for human activity measurement, action moment in games by replay frame in Convolutional Neural Networks (CNN), or possing the posture and body behavior from skeleton key points estimation with OpenPose library built-in function followed task-driven multisigmoid models — as well classification testing stroke accuracy could potentially make it into modern automated systems. These methods are combined to provide an in-depth qualitative understanding into racket sports players which can aid gameplay, enhance coaching strategies, increase performance and engage with fans.