Research on the Technology of Fire Smoke Detection Based on Image Process
|School||Xi'an University of Architecture and Technology|
|Course||Signal and Information Processing|
|Keywords||Block Segmentation Self-Adaptive Wavelet Transform Optical Flow Track|
The fire smoke detection technology based on image is not mature enough, proneto occur false or leakage alarm in practical applications at present. For the problems ofimage fire smoke detection in real time and accuracy, the paper put forward a newalgorithm from study of the following aspects:Firstly, common algorithms of suspicious area image segmentation were analyzed;the advantages and disadvantages of each algorithm were compared in the segmentationeffect, complex rate and so on. A image segmentation algorithm based on image blockand background adaptive updating was proposed, it could effectively overcome theimpact of light mutation and noise, and extract the moving targets completely.On the basis of motion detection, the color characteristics of smoke imagewere emphatically researched. The color model and brightness change model for smokewere respectively established in RGB space and HSV space. The static and dynamiccharacteristics of image smoke were researched, the edge irregularity, the blurringcharacteristics of background texture, smoke dispersion characteristics, smoke internalroll and the edge flashing frequency and so on were included, and the feature extractionalgorithms, such as edge complexity, high and low energy attenuation changerate,internal pixel changes in frequency and edge flashing frequency were designed. Thefeasibility and effectiveness of each feature extraction algorithm were verified throughthe simulation results. The fuzzy characteristic of the smoke on the background imagetexture based on two-dimensional discrete wavelet transform was mainly researched.Be aim at smoke movement characteristics, the Lucas-Kanade sparse light flowtracking algorithm based on Image pyramid layered was mainly researched. Firstly,good feature points suitable for tracking were extracted by Harris corner detection algorithm. The feature points of moving regions were continuously tracked trough theLucas-Kanade sparse light flow, then the real-time motion vector was obtained. threekey characteristic parameters: the average offset, movement trends and movementtrajectory were extracted to distinguish the difference between the smoke and interferce.Finally, the comparison experiment was given to prove the validity of the characteristicparameters.