Various Approaches of Human Activity Recognition: A Review

In the past few decades, recognizing the activities of an individual is remaining as the most challenging task in the computer science domain. Human activity recognition (HAR) is the study of identifying the actions of agents based on digital images and sensor data. Identifying human activity from digital images and sensor data is a considerably difficult task due to background disorder, partial occlusion, changes in scale, perspective problems, lighting, appearance, etc. To solve this problem, significant research has been performed in this area. Many applications of the HAR system includes video surveillance systems, human-computer interaction, etc. This work gives a concise survey on some new research advancements in the field of human activity recognition. Moreover, this work provides a comparative study on three popular approaches of activity recognition viz. vision-based (pose estimation), wearable devices, and radio signal-based approach. Also, it includes a comprehensive analysis on some existing datasets of HAR. The work includes a couple of pros and cons on the aforementioned approaches of activity recognition based on their performance and shows the research significance of vision-based system.