Analysis of Sensor Locations on Human Body for Wearable Sensor based Activity Classification during Fast Bowling in Cricket

This paper focuses on determining best body sensor position among calf, thigh, upper trunk and forearm when classifying Run Up, Delivery Stride and Follow Through phases during fast bowling in cricket by the usage of a machine learning model. Nine-axis Inertial Measurement Units (IMU) were used to collect data at 350Hz and Madgwick’s quaternion based algorithm was used for orientation estimation. The study also focused on determining best quaternion to be considered for such activity classification requirements in fast bowling. Three fast bowlers with Mixed type bowling action were considered for the study. A sliding window with 200 samples/window with 50% overlap collected eight, time domain statistical features from the sensor data and Principal Component Analysis was used to reduce dimensionality of the feature set. A linear kernel based Support Vector Machine classified the features into the three main phases and five-fold cross validation was used to determine model performance.