Human Activity Identification and Suspicious Behaviour Detection System

Abstract

Recognizing human action has been one of the biggest challenges in computer vision for the past two decades. Recently, it has become feasible to extract precise and cost-effective skeleton information. Our proposed system utilizes a cut-based framework to identify human actions using skeleton data. By using a single stationary camera as input, this system can recognize various continuous human activities in real-time, including raising or waving one or more hands, sitting down, and bending over. The recognition process is based on machine learning. Firstly, a dataset with the human body's coordinates is created. Then, a training model is developed using Logistic Regression and an outcome to be achieved. Finally, the model is utilized to identify human activities such as sitting, running, and waking up, as well as recognizing suspicious behaviour.