| Because of the limitation of the traditional fire detection, researchers at home and abroad begin looking for a new fire detection method in recent years, proposing a smoking detection method using video processing and digital image processing in the early of fire. Its advantage is that the scene can be monitored in real-time and the changes of smoke can be detected and tracked. Therefore, people can take effective measures to deal with it.Based on the inherent properties of smoke, this paper researches on the following three aspects of the smoke detection:(1)Build a background blurring model of smoke images based on two-dimensional discrete wavelet transform. The edge information of the image can be effectively extracted using two-dimensional discrete wavelet transform. A composite image is composed with the three sub-band structure gained by DWT. Then the blurring model is built by comparing the composite images of the current and the background image. In the results of experiment, the difference between the blurring model of smoke image and non-smoke image is showed.(2)Build the diffuse model of optical flow of the smoke image based on the quaternion wavelet transform. The optical flow field of smoke image can be driven using the phase information obtained from the coefficients of QWT. Then the diffuse model of optical flow is built based on the phase information of each optical flow. Finally the difference is presented by comparing the diffuse models of optical flow of smoke image and non-smoke image in the experiment results.(3)The smoke detection algorithm combined multiple properties. A union distinguish rule is presented based on blurring property, diffuse property and the property of primary orientation angle. Different measures are taken along with the different numbers of motion object in the image. Experimental results show that compared with the algorithm of detecting smoke using single property, the proposed method can improve the accuracy of smoke detection. |
PDF Full Text Request