Human Activity Recognition with Deep Learning: Overview, Challenges & Possibilities

Abstract

Human Activity Recognition (HAR) has become a vibrant research field over the last decade, especially because of the spread of electronic devices like mobile phones, smart cell phones, and video cameras in our daily lives. In addition, the progress of deep learning and other algorithms has made it possible for researchers to use HAR in many fields including sports, health, and well-being. HAR is, for example, one of the most promising resources for helping older people with the support of their cognitive and physical function through day-to-day activities. This study focuses on the key role machine learning plays in the development of HAR applications. While numerous HAR surveys and review articles have previously been carried out, the main/overall HAR issue was not taken into account, and these studies concentrate only on specific HAR topics. A detailed review paper covering major HAR topics is therefore essential. This study analyses the most up-to-date studies on HAR in recent years and provides a classification of HAR methodology and demonstrates advantages and disadvantages for each group of methods. This paper finally addresses many problems in the current HAR subject and provides recommendations for potential study.

Transformation classification of human squat/sit-to-stand based on multichannel information fusion

“…The pressure of the eight pressure points of the left and right feet in the three squatting movements is drawn as a curve as shown below. Figures 7,8,and 9 show that in the process of completing the three movements of half squat, full squat, and sitting, the pressure curves of the left and right feet are similar during the sitting-standing process. However, in the process of half squat-standing up and full squat-standing up, the pressure curves of the left and right feet are quite different.…”

“…[3][4][5] With the development of deep learning, the use of deep learning methods has surged, leading to an increase in recognition accuracy. [6][7][8] Many researchers have begun to study the application of deep learning convolutional neural networks (CNNs) and long short-term memory (LSTM) networks in RGB video action recognition. 9 Each method has its advantages in recognition, but it also has some inherent weaknesses.…”

Biometric Identification System based on Object Interactions in Internet of Things Environments

“…Some HAR systems recognize certain types of body movements by analyzing objects' orientation and acceleration [14,2,16]. Others precisely identify tasks such as cooking or cleaning the house by applying sensor fusion [12,11]. Using multiple sensor modalities improves the activity recognition performance and creates a better reference frame [9].…”

“…Therefore, a smart home should be able to distinguish who is interacting with these objects at a given moment to decide whether certain actions can be safely executed. While many existing solutions focus on the recognition of specific activities and gestures performed by the users of IoT devices [14,2,16,12,11], in this case, the attribution of those activities to specific household members is desired. Some studies look at identifying users using various on-device sensors [5,19].…”