基于嵌入式视觉的实时火灾探测系统的研究与开发

发布时间：2018-04-03 22:39

本文选题：火灾探测　切入点：近红外图像　出处：《江苏科技大学》2012年硕士论文

【摘要】：图像型火灾探测器是火灾探测器家族中的新秀，它融合了智能监控、图像处理、模式识别、嵌入式等先进技术为一体，拥有报警时间短、不受室内空间大小约束、支持室外环境、支持准确定位等传统探测器无法比拟的优势。目前图像型火灾探测器还未得到广泛应用，其主要原因在于，由于实际工况的复杂性，，火灾事件的图像探测技术目前仍处于实验阶段，还不够成熟，此外传统探测器已经形成成熟市场，对新型探测器形成较大的市场竞争，导致新型探测器较少提供给工程应用。本文提供了有关图像型火灾探测器的全套解决方案，主要做了以下几个方面的工作：（1）选择了效果较好且价格合适的相机加滤镜方式的红外图像采集方法，并给出了定制相机的CMOS图像传感器的选型方案；（2）结合现有的全局发光度算法，分析其缺点，在其基础上结合局部几何特征，研究了综合两类特征的火灾事件探测算法；（3）在可二次开发的智能工业相机上编写火灾事件探测程序，成功完成了计算机到嵌入式设备的移植，最终探测器可接入报警网络脱离计算机独立工作。（4）为加快处理速度，对算法做并行计算优化，以不变矩算法为例，在计算机平台上采用OpenMP/CUDA技术，将该算法的速度提升1.3-19倍。（5）通过实验得到大量数据，证实了算法的正确性、稳定性，测量了并行计算优化的实际效果。本文有以下几处创新点：（1）给出了一套基于局部几何特征的火灾探测算法，算法包括热源区域提取和匹配方法、5种红外热源区域几何特征计算方法、基于最小风险贝叶斯的多特征融合的热源区域分类方法、多重采样平滑判定噪声方法。（2）尝试将原创的基于局部几何特征的方法和文献[1]中的基于全局亮度特征方法结合起来，获得了速度快、效果好的优点，通过实验证明了整体效果优于文献[1]中的算法。（3）在计算机平台上对算法做并行计算优化，使用OpenMP/CUDA两种技术，给出完整的实现细节和优化效果。
[Abstract]:Image fire detector is a new star in the family of fire detectors. It combines intelligent monitoring, image processing, pattern recognition, embedded technology and other advanced technologies. It has short alarm time and is not constrained by the size of indoor space.Support outdoor environment, support accurate positioning and other advantages of traditional detectors can not be compared.At present, the image fire detector has not been widely used, the main reason is that because of the complexity of actual working conditions, the image detection technology of fire events is still in the experimental stage, and is not mature enough.In addition, the traditional detectors have formed a mature market, forming a greater market competition for the new detectors, resulting in the new detectors are less available for engineering applications.This paper provides a complete set of solutions for image-based fire detectors, mainly in the following aspects:1) the infrared image acquisition method with the better effect and the appropriate price is selected, and the selection scheme of the CMOS image sensor of the customized camera is given.(2) combined with the existing global luminosity algorithm, the shortcomings of the algorithm are analyzed, and based on the local geometric features, the fire event detection algorithm which integrates two kinds of features is studied.The program of fire event detection is written on the redeveloped intelligent industrial camera, and the transplant of computer to embedded equipment is successfully completed. Finally, the detector can connect to the alarm network and work independently from the computer.In order to speed up the processing speed, the algorithm is optimized by parallel computation. Taking the moment invariant algorithm as an example, the speed of the algorithm is increased by 1.3-19 times by using OpenMP/CUDA technology on the computer platform.5) A large number of data are obtained through experiments, which verify the correctness and stability of the algorithm, and measure the practical effect of parallel computing optimization.This paper has the following innovations:In this paper, a set of fire detection algorithms based on local geometric features is presented. The algorithm includes five methods for calculating the geometric features of infrared heat source region, including heat source region extraction and matching method.Multi-feature fusion based on minimum risk Bayesian classification method for heat source region classification, multi-resampling smoothing noise detection method.2) We try to combine the original method based on local geometric feature and the method based on global luminance feature in reference [1] to obtain the advantages of high speed and good effect. The experimental results show that the overall effect is better than the algorithm in reference [1].The algorithm is optimized by parallel computing on the computer platform, and the complete implementation details and optimization results are given by using the two techniques of OpenMP/CUDA.
【学位授予单位】：江苏科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;TP368.1

【相似文献】