Research On Flame Detection Based On Color And Motion Features

Fire is the one of the most dangerous disaster that seriously threaten public safetyand social development. The reasonable and effective control of fire is a significantindication of the progress of civilization. Researchers continuously generalize theroutine of fire, expecting to minimize the harm. If we can put the fire out in the earlystage of the burning process, because of low destruction of this stage, the losses wouldbe minimized. However, the capability and energy of human beings is limited, it’shard for us to detect the fire in the early stage of the burning process. So it remindspeople to develop an automatic fire detect system which can accurately and promptlydetect fire accident.Compared to traditional flame detection methods, making use of the increasinglymature computer vision technology to detect flame automatically is attracted manyresearchers’ view. In this paper, we suggest a color rule based model and a motionmodel, detecting the color features and motion features of the flame.Because the obvious color features of flame, the color based detection model isone of the most common methods. First of all, we analysis the characteristics of flamearea under the RGB color model and generalize the rules in the single channel,multi-channel and statistical models. In addition, We have constructed flame pixelsclassification model by making use of YCbCr color model. Besides the rulestransferred from the RGB or rgb model, we have develop the new rules based inYCbCr color model which could further alleviate the harmful effects of changingluminance. The flame pixel classification rates of the proposed new system iscompared with the previous method, and achieve a significant improvement.The motion features presented in the flame video sequences has indicated usmake use of this feature to detect flame. Because the object detection is based onflame detection, it’s different from the common objects like pedestrian or moving cars. So we has analyzed the advantages and disadvantages of the traditional methods, andwe has improved for the property of flame detection. Finally we proposed a newflame detection model based on motion features.In the experiment part of the paper, we has construct the process of the entireflame detection model, and the final flame detection system can be applied to realtime monitoring conditions. We have made use of candidate flame pixelsclassification method based on color model and improved moving target detectionmethod to construct detection model. Comparing with traditional flame detectionmodel, the model we proposed in this paper has achieved expecting effects.