Study Of Early Fire Prediction In Tunnel Based On Video

Fire detection based on video technology as a supplementary mean has beenmore and more focused on. Compared with the traditional fire detection technology, ithas a good real-time performance, strong anti-interference, visualization and low cost.In this paper, the detection technology of fire video in the tunnel was studied.The paper chose the fire events in traffic scene as research objects, aiming atdesigning and implementing the recognition algorithm of flame and smoke. Firedetection algorithm in this paper was mainly divided into the following part: theextraction and update of background, the location of the candidate area of smoke andfire, the analysis of the characteristic of flame and smoke, the design of classifierbased on Adaboost, the function and innovation of each part was explained as follow:1. The extraction and update of background: At first, the maximum and othervalue of the histogram was confirmed to be the confidence degree of the backgroundand selecting the higher degree of confidence to be the background based on thestatistic histogram method. In order to improve the reliability of the background, thebackground update algorithm based on block stability was used to update thebackground, using SAD as the measure criterion of similar degree between blocks ofbackground to update the weighted background.2. The location of the candidate area of smoke and fire: After obtaining reliablebackground, background difference method was adopted to detect the moving target,and then block binarization was used. In order to improve the efficiency of the laterprocessing, we preprocessed the selected moving target to exclude some interferenceregions, labeled the connected domains, eliminated small areas, and used the graystatistical consistency and movement accumulative characteristics to filter the targetaccording to the characteristic of the flame and smoke.3. The analysis of the characteristic of the flame: For the extraction of suspectedflame area, the characteristic of growth area, frequency band and circular degree offlame was mainly analyzed. 4. The analysis of the characteristic of smoke: For the extraction of suspectedsmoke, the height feature, texture feature and the main direction of the smoke weretook to analyze.5. The design of fire classifier: Each extracted characteristic of smoke and fire asa weak classifier was taken and Adaboost algorithm was used to fit the weak classifieras a strong classifier, so as to achieve rapid detection for fire.After taking different traffic scenes for experiments, the fire detection algorithmdesigned in this paper can be more accurate and effective to detect the fire events,which indicate that the algorithm has good reliability and applicability, and satisfy thereal-time requirement of the system.