Exploration of Human Activity Recognition Using a Single Sensor for Stroke Survivors and Able-Bodied People

Abstract

Commonly used sensors like accelerometers, gyroscopes, surface electromyography sensors, etc., which provide a convenient and practical solution for human activity recognition (HAR), have gained extensive attention. However, which kind of sensor can provide adequate information in achieving a satisfactory performance, or whether the position of a single sensor would play a significant effect on the performance in HAR are sparsely studied. In this paper, a comparative study to fully investigate the performance of the aforementioned sensors for classifying four activities (walking, tooth brushing, face washing, drinking) is explored. Sensors are spatially distributed over the human body, and subjects are categorized into three groups (able-bodied people, stroke survivors, and the union of both). Performances of using accelerometer, gyroscope, sEMG, and their combination in each group are evaluated by adopting the Support Vector Machine classifier with the Leave-One-Subject-Out Cross-Validation technique, and the optimal sensor position for each kind of sensor is presented based on the accuracy. Experimental results show that using the accelerometer could obtain the best performance in each group. The highest accuracy of HAR involving stroke survivors was 95.84 ± 1.75% (mean ± standard error), achieved by the accelerometer attached to the extensor carpi ulnaris. Furthermore, taking the practical application of HAR into consideration, a novel approach to distinguish various activities of stroke survivors based on a pre-trained HAR model built on healthy subjects is proposed, the highest accuracy of which is 77.89 ± 4.81% (mean ± standard error) with the accelerometer attached to the extensor carpi ulnaris.

Wearable sensors and machine learning in post-stroke rehabilitation assessment: A systematic review

“…Are systems which aim to identify specific movements of rehabilitation of the patients and differentiate between them for record and monitoring purposes [41][42][43][44][45][46][47][48][49][50][51], in this category researchers monitored Activities of Daily Living (ADL) [75] and they most frequently covered detecting general activities like standing, sitting, lying, standing up, sitting down [42,44,47,48,50], performing kitchen tasks like making a drink, chopping food [42] and other routine activities like making the bed, reading and lacing shoes [48], folding, sweeping and brushing teeth [46,48,49]. Other researchers covered activities for specific body parts like recognising different hand gestures [41], arm gestures [43] and some exercises to strengthen shoulders, and arms [48].…”

“…Some studies used individual accelerometers [43,45,53,54,65,67,68,70,71] or gyroscopes [45] while the rest used their combination to give more detailed information. Moreover, IMUs were coupled with different sensors to acquire more information: a barometric pressure sensor to detect changes in altitude [47,48], insole pressure sensors in [55] to measure the force exercised by the feet while performing the activities, flex sensors to measure the amount of deflection or bending while griping objects [68], liquid level detectors in a cup [42] to measure drinking activity and EMG sensors [61,51] to measure the activity of the muscles that can translate as strength. Only a single study did not use IMUs and employed EMG sensors only [41].…”

“…k-Nearest Neighbour (kNN) is another algorithm that was used in three included works [64,51,46]. The kNN classifier is based on distance metric and was widely used in real-time applications as it is free from the underlying assumptions about the distribution of the dataset.…”

“…The system aim is almost homogeneously divided between the three categories with eleven of the included works treating activity recognition [41][42][43][44][45][46][47][48][49][50][51], twelve dealing with movement classification [52][53][54][55][56][57][58][59][60][61]63,64] and ten with clinical assessment emulation. [65][66][67][68][69][70][71][72][73][74].…”

“…For the publication year a growing interest has been noticed starting from 2020 in the included sample of works. Two studies were published in 2015 [44,48], three in 2016 [56,68,69], four in 2017 [47,61,65,72], six in 2018 [41,42,46,52,59,66], five in 2019 [43,53,55,57,70] and then the maximum eight in 2020 [45,54,58,63,67,71,73,74] and finally five up to now in 2021 (up to August) [49][50][51]60,64]. Subsequently, publications made in a period of two years and a half (2019, 2020,2021) which consists of a third of the overall time range (seven years and a half) accounted for 55% of the total publications.…”