一种改进的帧差法实现火焰目标分割

针对传统船舶火灾报警系统采用的传感器在远离火灾源时很难探测到初期火灾的特性,提出了利用摄像头模式识别的方法来探测火源的构想,将顶置鱼眼摄像头作为图像传感器,在一个较大的范围内探测初期火灾,提出了一种基于火焰闪点位置的交叉性与聚合性的帧差算法,实现火焰目标分割识别.该方法是在实时视频流中,对连续两帧图像进行帧差,根据闪点位置的交叉性排除运动物体的干扰,再根据闪点的聚合程度排除一部分游离的干扰闪点,最后得到疑似火焰区域.对疑似火焰区域二值化、抽取轮廓、再由种子点位置确定火焰外形.此法能够高效地从视频图像中分割出火焰,从而为火灾的初期预防报警提供高效识别手段.

Flame segmentation by an improved frame difference algorithm

When the sensor of a traditional ship fire alarm system is far away from the fire source, it is generally hard to detect the fire early. This paper proposes a method of fire detection in a large region by pattern recognition, for which a fisheye camera is used as the visual sensor. A frame difference method based on the cross-cutting and polymerization for the locations of flame flash points is thus developed, in order to recognize the flame. By examining the difference of two consecutive frames in live video streaming, this method removes the interference of moving objects according to the cross-cutting of the flash point location, and then removes part of dissociative flash points according to the polymerization of the locations of flash points, and then obtains the suspected flame area. The profile of the flame can then be extracted by thresholding method. This method can be effectively used to get the flame segment from the images, which provides a recognition means for early detection and alarm of fire.