The Study Of Recognition Of Fire Smoke Video Images Based On Optical Flow And Multi-information Fusion Detection Algorithm

The most widely used fire detection technology nowadays, point-type photoelectric smoke detection belongs to contact and passive detection form which is spatial limited, so it cannot satisfy the fire detection requirement of the places like the large volume buildings, underground constructions, tunnel constructions, complex buildings. It is difficult for an infrared beam smoke fire detector to work well in large volume buildings, where the length of the protection space is over 100 meters or the airflow is too fast. In case of fire, if the fire source is far from the sampling point, it will cost much time for the fire signal from the fire source to the high sensitive aspirating smoke detector. Other fire detection technologies such as infrared and ultraviolet fire detection technologies are hard to be widely used as a result of high cost, because they need some very expensive optical equipment for spectral analysis.With the development of machine vision and computer technology, the video fire detection technology comes into being in the latest years, which has advantages, such as fast response, wide detection area, and little environment pollution. And video fire detection, which represents relatively high level of the fire detection technology nowadays, can be used for fire detection in large volume buildings, and even outdoor environment.Video fire flame image detection technology has already been widely used, but based on people’s priori knowledge, most fires begin with smouldering stage when smoke turns up earlier than flame, so that video smoke detection could realize very early fire detection. But the alarm failure and false alarm rate of video fire smoke detection are still comparatively high by now, so the study in this paper is about new method of video fire smoke detection, and some relative study about the arithmetic of reducing the alarm failure and false alarm rateThere is still no standard video database for fire video recognition. The video fire detection experimental system is built, and through recording the videos of tests, the representative fire smoke and disturbance characteristic video database is established, for the learning of the characteristic parameters and the better validating of fire smoke recognition model and the performance of video smoke real-time detection system. Some relative videos published online by foreign researchers are added to the video database for the comparison with other arithmetic.Several classic smoke image features and its extracting methods including color, blurring, texture and contour, are studied and analyzed in the paper. And by means of the fire video images library established, good point and bad point and applicable scenes of various smoke image feature extraction methods are analyzed. The real-time smoke image detection based on texture features is developed and realized at the same time.The application of optical flow method in video fire smoke recognition is studied. Using the optical flow method extract the optical flow of the feature points in smoke image regions. And the motion velocity field is established in order to reflect the varying trend of each gray pixel point in the image plane. The motion features and velocity field’s distribution characteristics of smoke images are analyzed. The velocity vectors of feature points are classified using neural network method, and a reliable feature model of the motion of smoke images is established, and the reliability and accuracy of video fire smoke detection systems is advanced.The laws of development of various fire characteristic parameters in the early of fire in confined space is studied, and the multi-information fusion algorithm combining smoke image information and smoke and CO concentration signal from sensors is developed, and the multi-information fusion fire detection in confined space is realized.The study of video fire smoke detection technology is still in the stage of development. The study of this paper will provide theoretical basis and technological support to the extensive use of video fire smoke detection technology.