System for Identifying Human Activities and Detecting Suspicious Behaviours

Suspicious activity identification from surveillance video is an effective research area of image processing and computer vision. Detecting suspicious activities is significant for maintaining the security of organizations and communities. Surveillance cameras are mostly used in public areas to monitor and secure safety. It is difficult to observe public places continuously hence intelligent video surveillance is needed that can detect human activity in real-time and classify them as non-suspicious & suspicious activities. By employing a stationary camera and integrating state-of-the-art machine learning algorithms—including Logistic Regression, Ridge Classifier, Random Forest, and Gradient Boosting—the system achieves real-time identification of diverse activities. The dataset, constructed from human body key points extracted through video analysis, demonstrates the system's efficacy in distinguishing between normal and anomalous behaviours. With potential applications spanning surveillance, healthcare, and public safety, this research underscores the system's capability to deliver timely alerts, thereby enhancing security protocols and minimizing reliance on manual monitoring. Keywords- Suspicious Activity, Non-Suspicious Activity Logistic Regression, Ridge Classifier, Random Forest, and Gradient Boosting.