Improving Basketball Recognition Accuracy in Samsung Gear S3 Smartwatch using Three Combination Sensors

Abstract

Development of Internet of Things (IoT) devices become popular to make it easier for people to recognize activity from wireless devices. Activity recognition has been widely used at various levels of computing. Smartwatch is one of IoT wearable devices used by researchers since its advantage for open source Human Activity Recognition (HAR) programming usage. Smartwatch in many published articles uses two sensors to accomplish HAR, which are accelerometer and gyroscope. However, the data obtained from the two sensors still too many restrictions in detecting sports activities such as basketball, football, and many more activities having an extreme movement. Moreover, previous experiments evaluate the impact caused by combining another sensor to get more precise of the activity recognition accuracy. Samsung Smartwatch Gear S3 has an audio sensor data that can be obtained from devices and have a promising result to improve recognition accuracy. This research proposed recognition accuracy by combining Accelerometer, Gyroscope, and Audio sensor to achieve improving accuracy from 69% become around 90% extreme movement recognition accuracy. The experiments show that Human Activity Recognition proposed is capable to detect Basketball activities on the Samsung Gear S3 smartwatch.