Search

Gaussian mixed model, LK optical flow method and background subtraction from foreground method are used to extract the fire and smoke region in foreground of video image. Multi feature of fire characteristics are used to extract the information. Colour feature of suspected region are extracted according to the colour model RGB and HSI spaces. Background blur feature is extracted using two dimensional discrete wavelet transform. If smoke appears in scene, the contour edge of the background would become blurry. The motion direction feature is extracted using LK optical flow method and gaussion mixed model. The DHT 11 digital temperature - humidity sensor in sensor node is used to extract temperature and humidity values for measurement and TIMSP430 micro controller for processing the information. The video node and sensor node extracted information are combined to detect the possibility of fire in the area during worst season conditions. By this method, the accuracy of fire and smoke detection is improved even in the worst environmental condition such as rainy weather. From the simulated and experimental results, the proposed method improves the accuracy and detection rate. Combination of sensor output and video output give excellent value in finding smoke or fire from videos. They reduces false detection rate of detecting smoke from non-smoke videos. It can be used in outdoor large environment.

Fire alarm system is based on detection of fire from video acquisition input data. This is done with the help of digital image processing techniques and embedded vision. It is based on vision-based fire detection system. This approach integrates colour, spatial, temporal and motion information to locate fire regions in video frames. The characteristics of burning flame and spatial temporal features of smoke combination is used to remove spurious fake fire regions. This video based fire alarm systems has stages such as frame conversion, fire and smoke colour detection, motion detection, spatial temporal analysis. Fire alarm will be activated when the system detects the occurrence of fire in a certain position and time interval surpasses the threshold.

This paper proposes an effective method to detect smoke in videos. The technique background subtraction is used to segment moving regions in fire video frame. The fuzzy c-means (FCM) method is used to cluster only smoke regions from these moving regions. Experimental results show that the proposed smoke detection algorithm performs well than the conventional smoke detection algorithms in terms of accuracy of smoke detection. It also provides low false alarm rate and high reliability in open and large spaces. The true positive value of all smoke frames and its percentage of overall smoke detection rate is more than 85%. The false positive value of non-smoke frames and its percentage of overall smoke detection rate is less than 10%. Hence overall smoke detection rate is increased.