基于视频图像的火焰多特征检测

发布时间：2018-04-19 17:04

本文选题：视频图像 + 颜色提取　；参考：《沈阳建筑大学》2014年硕士论文

【摘要】：火灾的发生一般会造成人员的伤亡和财产损失,因此,对火灾的准确的检测是人们关注的焦点。火灾探测器的出现标志着火灾探测的新时代,对人类的进步和社会的稳定有重要的意义。目前,在小空间和普通场所,传统的感温和感烟探测器占据着主导地位,其技术水平也已经很成熟。但是对于高大空间和存在干扰的场所,传统的探测器无法达到探测要求,急需一种有效的探测方式。随着图像识别技术和计算机技术的发展,基于视频图像的火灾探测技术应运而生。它具有快速响应、检测范围广、环保节能等的特点,与传统的探测器相比适合更多的复杂的环境,具有广阔的前景。目前,国内外也有一些成型的类似产品上市,但是因为其成本昂贵,所以普及困难。现有的视频火灾探测技术虽然种类繁多,但是仅靠单一的火灾静态特征或动态特征很难准确的对火灾进行检测。本文利用静态的颜色特征、纹理特征和动态的面积增长特征,再结合BP神经网络对火焰的识别,建立火灾探测系统。本文主要的工作和研究成果有以下几个方面：(1)颜色特征的提取方法有多种多样。本文在HIS色彩空间中对火焰颜色进行建模,对采集到的视频进行每秒4帧图像的获取,通过对视频截取的单帧图像进行颜色特征提取,获得疑似火焰区域。(2)利用边缘检测技术对疑似火焰区域进行分割,得到其边界。二值图像上的白色像素的个数认为等价于疑似火焰区域面积。然后在matlab平台上对3秒内的分割图像面积大小进行统计描点,利用线性拟合计算出一次函数的斜率,判断其大小。大于0,认为疑似火焰区域面积增大；小于等于0,认为疑似火焰区域面积减小或不变。(3)对图片提取其纹理和颜色通道的特征参量,建立BP神经网络。利用网上下载、自行拍摄和视频截取的220张图片的特征参量来训练网络,用80张图片来对网络进行测试,得到一个有效的神经网络。(4)总结颜色特征、火焰特征和BP神经网络的识别技术,提出了基于视频图像火焰多特征的火灾检测系统。利用matlab平台编写程序对其进行仿真实验。通过对11部视频的测试,认为系统效果较好。
[Abstract]:Fire usually causes casualties and property losses. Therefore, the accurate detection of fire is the focus of attention.The appearance of fire detector marks a new era of fire detection and is of great significance to human progress and social stability.At present, traditional temperature-sensing and smoke detectors occupy a dominant position in small space and ordinary places, and their technical level is also very mature.However, traditional detectors can not meet the detection requirements for large space and places where interference exists, so an effective detection method is urgently needed.With the development of image recognition technology and computer technology, the fire detection technology based on video image emerges as the times require.It has the characteristics of rapid response, wide detection range, environmental protection and energy saving. Compared with traditional detectors, it is suitable for more complex environments and has a broad prospect.At present, there are some similar products on the market at home and abroad, but it is difficult to popularize because of its high cost.Although there are many kinds of existing video fire detection techniques, it is difficult to detect the fire accurately by single static or dynamic fire characteristics.Based on the static color feature, texture feature and dynamic area growth feature, the fire detection system is established by combining the BP neural network to identify the flame.The main work and research results of this paper are as follows: 1) there are various methods for extracting color features.In this paper, the flame color is modeled in the HIS color space, the captured video is acquired by 4 frames per second, and the single frame image is extracted by color feature extraction.The edge detection technique is used to segment the suspected flame region and get its boundary.The number of white pixels on the binary image is considered to be equivalent to the area of the suspected flame area.Then the area size of the segmented image within 3 seconds is statistically described on the matlab platform. The slope of the first function is calculated by linear fitting and the size of the function is judged.If the area of suspected flame is larger than 0, the area of suspected flame is smaller than or equal to 0, and the area of suspected flame is reduced or invariable. 3) the feature parameters of texture and color channel are extracted from the image, and BP neural network is established.The network is trained by the characteristic parameters of 220 pictures taken by itself and captured on the Internet, and the network is tested with 80 pictures, and an effective neural network is obtained. (4) the color features are summed up.A fire detection system based on multiple features of flame in video images is proposed based on the recognition technology of flame and BP neural network.Matlab platform is used to write a program for its simulation experiment.Through the test of 11 videos, it is considered that the system is effective.
【学位授予单位】：沈阳建筑大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU998.1;TP391.41

【参考文献】