| Fire brings a great threat to human life and property security.Fire hazard can be greatly reduced if we can realize early warning of the fire.Considering that the flames and smoke generated by fire manifest a certain image features in the visible range,computer vision has been utilized to improve the efficiency of fire detection,and visual fire detecting technology has been developed based on image processing and pattern recognition.This dissertation is an attempt to provide a insight into this area,the main contributions are concluded as follows:Firstly,a new image-based fire detection algorithm is proposed,which is based on the fast classification for support vector machine.The fire video and suspected fire video are regarded as analysis objects.It provides a new method of classification characterized as self-learning which conquers the limitation of man-made setting thresholds in the traditional classification methods.Three characteristic values are selected to form three characteristic vectors which are variance ratio of flame areas,circularity and the number of sharp angles.The fast classification for support vector machine algorithm is utilized to train the fire identification model,classify the fire and suspected fire object with the trained classifier.The experiment results show that the number of selected support vectors has been reduced,and classification function is simplified as compared to traditional algorithm based on support vector machine.Moreover the speed of classification is apparently improved without loss of recognition precision.Secondly,a new fire detection algorithm based on a combination of color model and the sparse representation model is proposed.The overall feature is extracted instead of single features of the flame in this algorithm.Firstly,the suspected area from the prepossessed image is extracted by color modeling in the space of HIS.Then sparse representation model is introduced,and the training samples are utilized directly to form the dictionary of fire and suspected fire.At last,the minimum residual between testing samples and training samples is calculated via sparse decomposition and reconstruction to classify the flame and interference.The experiments results show that comparing to traditional methods,such as neural network and look-up table,the new algorithm improves the classification rate and reduces the complexity.Finally,for the purpose of practical application of the fire detection in the subway command center,the implementation approach of visible fire detection based on the distributed structure is proposed.An algorithm of fire flame detection based onmuti-feature fusion,an algorithm of smoke detection based on L-K sparse optical flow method and an algorithm of fire spatial orientation based on multi-point detection are designed for the large space building.A distributed visual fire alarm and linkage system using DSP and FPGA has also been designed and implemented.The overall design is put forward;meanwhile the implementation scheme of the software and hardware hierarchical design is also presented.The multi-point detection system is adopted in the proposed scheme,including distributed processing,central processing and control.Video capture and detection are implemented by independent video detection and control unit,so it not only improves processing ability,lower power consumption and network group ability,but also solves problems generated by centralized data processing.It also promotes intelligence level of automatic fire protection system by combining detection equipment with fire extinguishing equipment. |
PDF Full Text Request