REAL TIME OBJECT NEGLECT DETECTION USING DEEP LEARNING AND VISUAL ATTENTION ANALYSIS

Abstract

Real-time surveillance systems are increasingly deployed in public environments such as airports, railway stations, shopping malls, and educational campuses to enhance safety and security. However, continuous manual monitoring of surveillance cameras is inefficient and prone to human error due to the massive volume of video streams generated every day. One critical security challenge in such environments is the detection of unattended or neglected objects such as bags, suitcases, or suspicious packages that may pose potential threats. This research presents a real-time object neglect detection system that utilizes advanced deep learning techniques and visual attention analysis to automatically identify unattended objects in surveillance footage. The proposed system integrates the YOLOv8 object detection algorithm for accurate identification of persons and objects and the DeepSORT tracking algorithm for consistent multi-object tracking across video frames. A distance-time based association mechanism is implemented to determine the relationship between detected persons and objects. When an object remains beyond a predefined spatial distance from its associated person for a specified duration, the system classifies it as a neglected object and generates an alert for security personnel. Additionally, the system incorporates explainable artificial intelligence (XAI) techniques such as saliency maps and distance timeline visualization to provide transparency and interpretability of the detection process. The proposed framework improves surveillance efficiency by reducing manual monitoring requirements while enhancing detection accuracy and reliability. Experimental analysis demonstrates that the system effectively detects unattended objects in real time, thereby supporting proactive security measures in smart surveillance environments.