Intelligent real-time monitoring of tunnels and fire early warning based on multisensors

As a key transportation node, the safety of tunnels is critical for people and vehicles. Traditional monitoring methods suffer from incomplete information, poor real-time performance and low intelligence. Therefore, in order to improve the accuracy of tunnel environment monitoring and the real-time of fire early warning, an intelligent real-time monitoring and fire early warning scheme based on multi-sensor data fusion and machine vision. The system integrates temperature, CO, and CO2 sensors with high-resolution visible-light and thermal cameras to capture spatial and dynamic fire indicators. A hybrid convolutional neural network (CNN) combined with bidirectional long short-term memory (Bi-LSTM) layers processes visual data, enabling flame and smoke detection in complex tunnel environments . Multi-modal fusion using Dempster-Shafer evidence theory further enhances reliability by combining sensor data with visual probability assignments. Experimental results indicated that the temperature sensor had a maximum error of only 0.87°C in the range of 0°C-25°C, and the carbon monoxide and carbon dioxide sensors also demonstrated high accuracy. The backpropagation neural network converged at the 23rd iteration on the training set with a loss value of 0.06, The integration of machine vision reduces false alarms caused by non-fire thermal anomalies (e.g., vehicle exhaust), achieving a detection accuracy of 99.3% and a false alarm rate of 1.2% in simulated fires. The fire early warning model attains an average response time of 8.2 seconds, significantly outperforming manual methods (29.86 seconds). This holistic approach provides accurate environmental monitoring, rapid fire response, and enhanced intelligence for tunnel safety management.