【作者】 金顺敬；

【导师】 汤一平；

【作者基本信息】 浙江工业大学 ， 计算机应用技术， 2008， 硕士

【摘要】 本文在图像获取手段上引入全方位视觉,在图像的分析、理解和信息加工上引入以人为中心的思想进行系统设计,利用计算机视觉技术,集动态图像理解和多特征信息融合于一体,对早期火灾的智能探测进行了探索,尝试实现一种能真正适用于大空间的图像型火灾安全监控系统。研究表明本文提出的系统能解决现有图像型火灾安全监控系统存在的监控范围小、实时处理差、实现算法复杂、对环境依赖性强等不足。本文的主要工作和成果总结如下:1.把实验室自主研发的全方位视觉传感器应用到火灾安全监控,并搭建嵌入式视频服务器,实现对大空间的实时全景监控。实现了全方位图的柱状全景图展开算法,使后续图像处理中的算法能够正确、方便的应用。2.引入以人为中心的思想,利用动态图像语义理解的理论,设计了一种面向对象的火灾探测模型。该模型通过分析火焰对象的属性实现视频序列中火焰对象的识别,通过分析火焰对象的行为实现火焰燃烧模式的识别,并做出初步的火灾判断。为火灾安全监控系统的设计开发提供了一种新的思路。3.研究普通CCD摄像头采集的视频序列中火焰目标对象的识别,详细介绍了在像素级使用运动、颜色和闪烁频率特征抽取火焰像素点,并实现了基于轮廓遍历的连通区域填充算法确定整个火焰区域,初步完成了具有语义信息的火焰目标对象的识别。4.研究火焰面积增长、整体移动和边缘抖动等行为特征的量化分析算法,并应用模糊逻辑实现多特征融合,设计实现了基于模糊逻辑的火焰燃烧模式识别算法,把火焰的燃烧区分成环境中可能存在的可控燃烧和会引发火灾的失控燃烧两种模式,为早期火灾探测做好了充分的准备。5.使用Java语言设计实现了火灾安全监控系统,实验了多种光照条件下的火灾安全监控,各种实验结果表明本文提出的大空间的火灾安全视频监控系统,具有监控范围大、实时处理好、实现算法精炼、对环境依赖性小等优点,通过多特征融合技术实现火灾判断的检测率有所提高,误检测率大大减少。虽然本文对图像型火灾安全监控技术做了一些基础性的研究,但离图像型火灾探测的实际应用还有一定的距离,还有很多内容需要进一步的深入研究,希望本文的工作能够对后来的研究者起到一定的参考作用。更多还原

【Abstract】 This paper introduced omni-directional vision to achieve a fire security system truly applicable to large space fire monitoring. According to the way people understanding the fire, this system used computer vision technology, combined with dynamic image understanding and multi-feature fusion, to implement intelligent fire detection. Experiments showed that this system could solve the shortcomings of existing vision based fire security system, such as poor real-time processing, complex algorithm, dependent on environment, etc.The main work and the results were summarized as following:1. Introduced omni-directional vision sensor to the fire security system, and structured embedded video server to achieve real-time panorama monitoring of large space. Developed unwrap algorithm to get panoramic image from omni-image, which was much better human readable and could be used properly in subsequent image processing.2. Designed an object-oriented fire detection model based on dynamic image understanding theory. This model achieved flame recognition by analyzing the properties, understood flame combustion pattern by analyzing the behavior, judged fire based on these information. This approach provided a new way on designing an intelligent fire security system.3. Researched on flame recognition of video sequence acquired by ordinary color camera. Detailed fire pixels extraction in the use of movement, color and frequency features. Designed and implemented the connected component filling algorithm based on contour traversal to identify the flame target region with semantic information.4. Researched on quantitative analysis algorithm of flame behaviors such as area changes, overall spread and contour characteristics. Applied fuzzy logic to achieve multi-feature fusion, and then designed the flame combustion pattern recognition algorithm which divided it into two different patterns: controlled combustion pattern and out of control fire burning pattern.5. Designed and implemented the fire security system using Java programming language. Performed a variety of experiments under different conditions of illumination, and achieved good experimental results. It showed that multi-feature fusion technology could increase the fire detection rate significantly. Although this paper has done some basic research on vision based fire security technology, there is still a certain distance to the practical application of fire detection. Hope this work would be able to play a certain role in the reference to the further researchers on the same topic.更多还原