基于视频监控平台的室内火焰探测算法研究

发布时间：2018-01-10 09:02

本文关键词：基于视频监控平台的室内火焰探测算法研究　出处：《天津大学》2014年硕士论文　论文类型：学位论文

【摘要】：火在人们的日常生活中扮演着重要的角色,但如果发生火灾,后果将是灾难性的,为了减少火灾造成的损失,要在火灾初期尽可能发现火情,将火灾消灭在萌芽状态,而现有的火灾报警设备在实际使用中存在一定的局限性,容易受外界环境的干扰,不能实现火灾初期报警。为了克服现有火灾探测器的种种缺陷,本文提出了一种基于视频监控平台的室内火灾探测算法,能有效克服现有火灾探测设备的缺陷,在火灾发生的初期发出报警信号,将火灾造成的损失降到最低。本文研究基于现有视频监控平台,通过SDK开发包获取视频码流,将获取的视频码流转化为便于进行视频处理的格式,同时采用共享内存的方式实现不同进程间视频数据的共享。首先,研究了火焰前景提取算法,包括帧差法、高斯背景建模法和光流法,在深入研究这些算法的基础上,结合火焰的运动特征和色彩信息,提出了一种新型的火焰前景提取算法,通过实验验证此前景提取算法能够得到完整的火焰轮廓,便于后续火焰特征的提取。在得到完整火焰轮廓的基础上,深入研究了在不同场所、不同光照强度下火焰识别的主要特征,包括静态特征和动态特征,以及火焰特征的提取算法,最终决定采用颜色特征、频率特征、相关性特征、圆形度特征、向上性特征等作为火焰识别的主要特征。最后,通过大量实验验证了本系统能准确识别出各种不同室内环境下突发的火焰,同时能有效地避免光线移动、人影晃动所造成的干扰。实验结果表明,本文算法的识别准确率在90%以上,识别时间大约在5-20秒。
[Abstract]:Fire plays an important role in people's daily life, but if there is a fire, the consequences will be catastrophic, in order to reduce the loss caused by fire, may be found to do fire early in the fire, the fire will be nipped in the bud, and the existing fire alarm equipment has certain limitation in practical use that is easily disturbed by the environment, can not be achieved at the initial stage of a fire alarm. In order to overcome the shortcomings of existing fire detectors, this paper proposes a fire detection algorithm based on video surveillance platform, can overcome the existing fire detection equipment defects effectively, an alarm signal in the early stage of the fire, the fire loss to a minimum. This paper based on the existing video monitoring platform, access to video stream through the SDK development kit, the video stream into easy video processing and format. By the way of shared memory sharing video data between different processes. First, study the flame foreground extraction algorithm, including frame difference method, Gauss background modeling method and optical flow method, based on in-depth research on these algorithms, combined with the motion of the flame features and color information, proposes a new algorithm for extracting foreground of flame through the experiment, this algorithm can extract the foreground of flame contour complete, easy to extract the features. In the subsequent flame based complete flame contour, studied in different places, different light intensity of the main features of flame recognition, including static and dynamic feature extraction algorithm, and flame characteristics, finally the color characteristic, frequency characteristic, correlation characteristic, circularity, to feature as the main feature of fire flame recognition. Finally, through a lot of experiments. The system can accurately identify all kinds of unexpected fires in indoor environment, and effectively avoid the interference caused by the movement of light and the motion of human figures. The experimental results show that the recognition accuracy of the algorithm is over 90%, and the recognition time is about 5-20 seconds.

【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN948.6

【参考文献】