A Comparative Study of Various Human Activity Recognition Approaches

• Published in: IOP conference series. Materials Science and Engineering Vol. 1131; no. 1; p. 12004
• Main Authors: Goel, Dhruv, Pradhan, Rahul
• Format: Journal Article
• Language: English
• Published: Bristol IOP Publishing 01.04.2021
• Subjects: Accelerometers; Artificial neural networks; Automatic control; Comparative studies; Data collection; Health care; Human activity recognition; Human motion; Machine learning; Motion perception; Moving object recognition; Scientific papers; Surveillance; Three dimensional models
• ISSN: 1757-8981; 1757-899X
• DOI: 10.1088/1757-899X/1131/1/012004

Human Activity Recognition (HAR) is a vast and exciting topic for researchers and students. HAR aims to recognize activities by observing the actions of subjects and surrounding conditions. This topic also has many significant and futuristic applications and a basis of many automated tasks like 24*7 security surveillance, healthcare, laws regulations, automatic vehicle controls, game controls by human motion detection, basically human-computer interaction. This survey paper focuses on reviewing other research papers on sensing technologies used in HAR. This paper has covered distinct research in which researchers collect data from smartphones; some use a surveillance camera system to get video clips. Most of the researchers used videos to train their systems to recognize human activities collected from YouTubes and other video sources. Several sensor-based approaches have also covered in this survey paper to study and predict human activities, such as accelerometer, gyroscope, and many more. Some of the papers also used technologies like a Convolutional neural network (CNN) with spatiotemporal three-dimensional (3D) kernels for model development and then using to integrate it with OpenCV. There are also work done for Alzheimer’s patient in the Healthcare sector, used for their better performance in day-to-day tasks. We will analyze the research using both classic and less commonly known classifiers on distinct datasets available on the UCI Machine Learning Repository. We describe each researcher’s approaches, compare the technologies used, and conclude the adequate technology for Human Activity Recognition. Every research will be discussed in detail in this survey paper to get a brief knowledge of activity recognition.