Smart Phone ROM Measurements Based On Action Recognition In Rehabilitation Using Deep Learning

Abstract

Computer-assisted physical therapy (rehabilitation) assessment includes assessing patient performance in undertaking recommended recovery tasks, based on interpreting movements from activity which are recorded through a sensory interface. For assessing rehabilitation as well as for determining permanent disability scores in workers' compensation situations by evaluating Range of Motion (ROM) following injury or action is needed. Smartphone technology developers have built software (apps) in recent years that imitate ROM measurement instruments, such as the universal goniometer. This paper explores the approaches to Activity Recognition (AR) using a smartphone application that utilizes the on-board accelerometer sensor with the goal of tracking people's physical activity at home. In addition, HAR may provide useful information about the level of daily physical activity via smartphone or wearable sensor, especially in circumstances where a physically AR are normally exists, as in modern environments. This paper suggest a deep convolution neural network system (DCNN) using a smartphone ROM measurement of various joints, such as leg, foot and ankle based on AR. Datasets are collected from android phones for the exercise movements done by patients and predict 88% recovery score. The proposed DCNN method obtained the highest training rate of 94.08% and higher testing accuracy of 94.17% than the LSTM method. DCNN method has spends 87sec less computation time.