Ai-based Surveillance System For Abandoned Object Detection Using Yolov8

Abstract

The Proliferation Of Surveillance Infrastructure Across Transportation Hubs, Commercial Complexes, And Civic Spaces Has Not Been Matched By A Proportional Improvement In The Capacity Of Human Operators To Effectively Monitor Multiple Video Feeds Over Extended Durations. Cognitive Limitations Inherent To Sustained Visual Monitoring Create Critical Gaps During Which Security-relevant Events, Including The Placement Of Unattended Items, May Go Unnoticed. This Paper Proposes A Computationally Efficient, Learning-based Framework That Autonomously Identifies Abandoned Personal Belongings Within Live Surveillance Footage By Integrating The YOLOv8 Single-stage Detector With A Spatiotemporal Ownership Inference Mechanism. The System Processes Individual Video Frames To Simultaneously Detect Persons And Personal Items — Encompassing Backpacks, Handbags, Laptops, And Mobile Devices — Using COCO-pretrained Weights Applied In A Zero-shot Configuration. Temporal Continuity Is Preserved Through An IoU Matching Strategy And Ownership Attribution Is Determined By Euclidean Proximity Analysis. Should The Attended Condition Remain Unsatisfied Beyond A User-defined Temporal Threshold, The Item Is Reclassified As Abandoned, Prompting A Visual Alert. Experimental Validation Confirms Near-real-time Throughput And Reliable Detection Outcomes.

Keywords