Bowling action recognition using computer vision

In the realm of computer vision with respect to cricket domain, object tracking with precise pose prediction has proven to be difficult and hence less work is found in this region. Object detection is different from such applications, as used for identifying the type of object present, while the task of tracking and identifying the unique identity of the specific object which is detected is one of the main objectives. Object tracking technology has been found to have a wide spectrum of applications in the sports industry and can be used to identify different types of athletes based on their unique characteristics of action performed while bowling. This study focuses on identifying cricket players and distinguishing between bowling and non-bowling actions based on the various poses they adopt while playing the game. Yolo V3 is the object detection system, here it is used to detect cricket players while MoveNet lightning is used for the detection of keypoints of the individual. The CNN acts as a classification model, which is designed for the classification of action, which is identifying each athletes pose and enabling more accurate classification of bowling or non-bowling action.