A Hidden Markov Model for Detection and Classification of ARM Action in Cricket Using Wearable Sensors

Hidden Markov Models (HMM) have been used for to accurately model, detect and classify key phenomenon. In this manuscript, we propose use of HMM for detection and classification of arm action in the game of cricket. The technique uses sensor data gathered from wearable sensors placed at wrist, elbow and shoulder. The sensor data consists of both displacement and rotational information collected through a combination of accelerometer and gyroscope placed at each joint. A Bluetooth transceiver is attached to the arm in order to wire-lessly transfer the gathered data to the base station. A K-means clustering algorithm classifies the current position and angular rotation of the joint for each of the sensor placements. A Markov chain then determines the chain of sequence for a set of joint movements (displacement and angular rotation) to classify it as a specific arm motion. A Hidden Markov Model determines the previous state of arm motion in order to classify the cur-rent state and hence, the current action since the movements happen in progression, when following the other. Experiments show an accuracy of up to 100% in correctly determining the arm action against a model built around a trace-set collected from a sports biomechanics expert. The proposed model has applications in cricket coaching and technique adaptation both for novice and trained players.