Enhancement of the Infrared Imaging Detection Model for Highway Tunnel Fires Using the AdaBoost Algorithm

Abstract

In order to effectively detect the flame in the early stage of highway tunnel fire and issue early warning, we proposed an innovative flame identification method to solve the problem of slow response speed of traditional temperature-sensing fire flame detector in large space environment such as highway tunnel. Based on the multi-feature and AdaBoost algorithm of the flame image, the static and dynamic features of the runaway flame in the image are studied, the motion foreground is extracted by the inter-pause difference algorithm, and the suspected flame area is segmented according to the color statistical model of the flame in RGB and Lab space. We further extract the eigenvalues of the first-order moment, circularity and LBP first-order moment of the H component from the suspected region to form the input vector of the AdaBoost static feature model, and construct the AdaBoost comprehensive feature model by combining the dynamic features such as the normalized eigenvalue of the beat frequency of the flame centroid and the proportion of the flame ton. In order to verify the effectiveness of the proposed method, we used the highway tunnel flame video and public video to conduct rigorous experimental tests on the trained AdaBoost static feature classifier and comprehensive feature classifier. The method achieves high-precision initial flame identification in the highway tunnel environment, with an accuracy of 97.21% and an accuracy of 98.01%. In addition, we provide statistics such as confidence intervals to support the accuracy of the report, and demonstrate a benchmark against state-of-the-art methods in the field, demonstrating that the method has significant advantages in eliminating false alarms caused by false flame interference.