Experimental Research on Intelligent Monitoring Method for Tunnel Water Inrush Based on Target Detection

To address the issues of insufficient coverage and poor real-time performance in monitoring tunnel water-inrush precursors,an unattended monitoring system based on video motion features was developed.Optical flow,frame difference,and background subtraction methods were evaluated using a laboratory physical model and field video data from the Shizishan Tunnel of the Central Yunnan Water Diversion Project.Performance was assessed regarding inrush-point localization and identification of changes in discharge under four conditions:a single inrush point with constant discharge (without dust and fog),an increase in the number of water-inrush points,an increase in discharge,and dust/fog interference.Results indicate that all three methods are feasible for water-inrush identification.The optical flow method exhibited robust performance with low sensitivity to foreground-background color contrast;however,it processed 38 700 frames at only 5~6FPS (approx.1h59min),failing to meet real-time requirements,and was susceptible to specular reflections,leading to incomplete recognition.The frame difference method had a low computational cost and ran at 80~85FPS (approx. 8min) but was sensitive to transparent or low-contrast water flows and dust/fog,often resulting in ghosting and contour loss.The background subtraction method achieved the highest speed of 105~115FPS (approx. 6min) and was relatively insensitive to illumination changes,yet its performance degraded when the background changed frequently or image quality declined.This study clarifies the trade-off between robustness and real-time efficiency for these methods in tunnel water-inrush monitoring.