Yolo-minesafe: A Vision-based Abnormal Fall Detection And Emergency Alert Framework For Isolated Mining Workers

Abstract

Mining Operations Consistently Rank Among The World’s Most Hazardous Occupational Environments, With Workers Stationed In Isolated Areas Facing Undetected Fall Risks, Sudden Health Emergencies, And Life-threatening Incidents That Current Safety Systems Cannot Address In Real-time. Existing Solutions, Such As Passive Closed-circuit Television (CCTV), Wearable Accelerometers, And Manual Supervision, Fail To Deliver Autonomous, Real-time Incident Detection Across The Expansive And Harsh Terrain Of Active Mine Sites. This Study Introduces YOLO-MineSafe, A Vision-based Fall Detection And Emergency Alert Framework Purpose-built To Close This Gap. The System Continuously Processes Surveillance Camera Videos Using A Fine-tuned YOLOv8 Deep Learning Model, Extracting Bounding Box Geometry, Posture Orientation, And Inter-frame Motion Vectors To Identify Anomalous Body Positions. A Temporal Classification Module Employing A 20-frame Confirmation Window At A 0.4 Confidence Threshold Reliably Distinguished Genuine Fall Events From Ordinary Work Postures, Such As Bending Or Crouching. Upon Confirmed Detection, Multichannel Emergency Alerts Are Dispatched Immediately: An Annotated Incident Image Via Email, An SMS To Registered Supervisors, And A Simultaneous Local Audio Alarm — All Without Human Intervention. The System Operates Effectively In Low-light And Dust-heavy Environments Through Dedicated Preprocessing, Requires No Wearable Devices, And Provides A Complete Incident Audit Trail, Representing A Substantive Advance Toward Reducing Preventable Fatalities In Isolated Mining Environments.

Keywords