Research And Application Of Smoke And Fire Detection Algorithm In Intelligent Surveillance System

The emergency, high frequency and strong destructiveness of fire, which threat people’s life and property seriously, thus fire prevention and detection has become an urgent issue in today’s society. The current widely used fire warning systems based of sensors are vulnerable to environmental impact. Under the complex environment of large space,the warning of system is not timely, and the rates of false results is higher. The fire detection algorithms based on video is not affected by space and distance, and it has a large range of detection and could meet the requirements of real-time detection system. In order to improve the reliability and reliability of video fire detection system, this paper mainly studied the video of flame and smoke detection technology The content of this paper include following aspects:(1) Pre-processing. Pre-processing operations consist of image smoothing and moving object segmentation. This paper has compared and realized four kinds of image smoothing algorithm, including mean filter and its improved algorithm, median filter and its improved algorithms. The paper has verified that adaptive median filter algorithm in noise reduction and preserving details of original image has the best effect. In terms of moving object segmentation, the paper has realized the inter frame difference and background difference method, and background difference method is better in the segmentation of foreground objects.(2) The flame detection. Through static and dynamic characteristic analysis of video flame, this paper has put forward and realized a scheme that use mean background subtraction divide the motion area as a mask,and combine with the feature of flame color and stability of motion accumulation to find out the suspected flame area. At the same time, the detection of suspected flame in boundary shape of region and regional similarity verify that the fire detection system can meet the requirement of real-time and accuracy.(3) The smoke detection. Through the feature detection algorithms of the fuzziness, float and texture characteristics, the scheme can output a stable smoke region in shortly time. Firstly, as the fuzzy feature of smoke image has the feature that its high frequency energy is decreasing. This paper through the wavelet transform of the smoke image, extract the change characteristics of high frequency energy in smoke image. Secondly, the paper has improved an existing fast motion estimation algorithm, the results show that it expand the scope of searching and matching, make the matching result is more accurate and realize the flutter characteristics of smoke extraction better. Finally, as the phenomenon that the smoke makes the covered background regions be smoother, the paper proposes a texture feature extraction method to detect the change of the edge, and it has a high detection efficiency.Finally, in this paper, the video detection algorithms for flame and smoke detection are simulated and tested, the test results show that the smoke and flame detection system could reach the requirements in the real-time, accuracy and robustness. Two dimensional image can not provide direct fire information of the actual size. Combining the further need, proposed an optimization scheme of the system, and verify the scheme.The scheme which is implemented under the camera calibration can detect the actual size of the fire, and conduct the fire severity classification, the classification information can be used to make fire emergency plan for The fire department.