DeepSecure – Suspicious Human Activity Recognition From Surveillance Videos

Abstract: The increasing deployment of surveillance systems has created a need for intelligent monitoring solutions that can automatically interpret visual data instead of relying solely on manual observation [1], [4]. DeepSecure - Suspicious Human Activity Recognition from Surveillance Videos addresses this challenge by integrating deep learning techniques with computer vision to detect abnormal or potentially dangerous activities in real time [1], [10]. The system is designed to identify suspicious human behaviours such as violent actions, panic movements, and unauthorized gatherings, while simultaneously detecting environmental hazards including fire and smoke [5], [6]. Using convolutional neural networks and YOLO-based object detection, DeepSecure effectively analyses both spatial and contextual information from live and recorded video streams [1], [10]. A Flask-based web application enables users to interact with the system through a browser interface, supporting both live camera feeds and uploaded surveillance footage [4], [10]. OpenCV is employed for efficient video processing, and a MySQL-backed authentication mechanism ensures secure access control [4]. The modular design of the system allows flexible deployment across public and industrial environments such as airports, educational institutions, and smart-city infrastructures.