Intelligent Video Surveillance The Pyrotechnic Detection,

Fire is one of common natural disasters,it caused huge losses to people's lives and property. If a fire alarm is in time, the losses brought by fire will be reduced significantly. Therefore, design an efficient algorithm of the flame detection and develop an appropriate fire detection system has a great significance to people's security and property. Recently, with the rapid development of the computer vision and pattern recognition, more and more attention has been paid to fire detection based on intelligent video surveillance and it also has been become a hot research topic in the field of computer vision.Most of the existing fire detection system based on video using a number of characteristics to identify the flame, such as color, shape, size change. However, due to the fire's produce is not regular and the change is complex, a high false alarm rate exists in current recognition algorithms and they can not suit for complex scenes, so they can't be used in large-scale industrial applications. How to use the flame's essential characteristics and to establish a reliable model to describe these features has become a challenging task for fire detection.Through careful analysis of the important features of the flame and combine the related theory of the computer vision and pattern recognition, in this thesis, an algorithm of hybrid flame detection based on spatial-temporal features of the hidden Markov model (HMM base on spatial-temporal features )and a saliency map based on the brightness has been proposed. The algorithm first get the candidate flame region through the motion region detection and color characteristics analyze, then combined with Hidden Markov Models (HMM) and visual attention saliency map model of the brightness to detect fire. The HMM measures the flame's flickering characteristics and the brightness saliency map character the brightness variation of the local area of the flame respectively. At last, the flame detection algorithm has been judged whether a suspicious motion area is a true fire region or not. A large number of experiments show that our method is new flame detection method with a high detection rate and low false alarm rate. Compared with existing flame detection algorithms, our algorithm has obvious detection advantages.Finally, we also implemented the smoke detection system prototype based on intelligent video surveillance. The method first get the candidate smoke region by the motion region detection and gray histogram analyze, then the wavelet transform are used for describe the high frequency energy of the smoke region and background region on candidate smoke region. The algorithm through excludes the moving objects looks like smoke and implement the smoke detection ultimately.