A Dual-Stage Hybrid Vision Framework Using YOLOv8n-Canny Edge Detection for Real-Time Railway Trespassing and Intrusion Monitoring

Abstract

Intrusion and trespasser detection on railway tracks is a crucial safety measure to prevent accidents and maintain operational reliability. This study proposes a hybrid vision-based approach that integrates YOLOv8n, a lightweight real-time object detection model, with the Canny edge detection algorithm to identify and classify unauthorized objects and individuals on railway tracks. In this context, intrusions refer to inanimate objects such as rocks, fallen trees, or construction materials obstructing the tracks, whereas trespassers refer to humans or other living beings engaging in unauthorized activities near or on the railway line. YOLOv8n is employed as a single-stage detector to localize and classify objects, while Canny edge detection is applied to enhance object contours and improve shape-based differentiation between intrusion and trespasser categories. Experimental results show an average accuracy of 52.37%, indicating moderate detection performance. Although the accuracy remains limited, the findings demonstrate the potential of combining deep learning and traditional image processing techniques to develop an automated monitoring system that supports railway safety and surveillance applications. Further optimization of the dataset, model tuning, and feature enhancement are recommended to improve detection performance.