Computer Vision based Activity Recognition: Studying and Chit chatting

Abstract

In a tech-driven and automated world, it is no surprise that innovation and research are reaching new heights and one such progress that has been researched, falls in the domain of image recognition, that is, Human Activity Recognition (HAR). Recent studies have shown their interest in human activity recognition systems that use computer vision and various machine learning algorithms to classify an image into different activities over which the model has been trained. A detailed review of many existing similar literature works that follow the CV-based projects for recognition purposes was also referred. This paper mainly focuses on activity recognition of studying and chitchat activities. The proposed method for HAR is purely CV based which uses a dataset of 2400 images, equally divided into two different activities, containing 1200 per activity. This work will be beneficial in recognizing human behavior, in surveillance and assisted living, elder care, and healthcare monitoring systems along with trending research areas such as human-robot interactions as well as gaming and entertainment. The best results were showcased by the KNN classifier after using the BRISK feature detector achieving an overall accuracy of 78 percent.