基于低照度室内视频图像的烟雾检测算法研究

发布时间：2018-06-17 06:34

本文选题：低照度 + 室内　；参考：《西安科技大学》2017年硕士论文

【摘要】：基于视频图像的烟雾检测是消防领域中的一种新型火灾探测技术。针对视频烟雾检测,目前国内外的研究大多在光线较好的环境下,低照度条件下的研究虽然还处于一个萌芽阶段,但是对火灾的预防却有着十分重要的指导意义。低照度条件下的火灾不但造成更多的人员伤亡,而且造成民众的财产损害也是极其严重的。本文的主要工作是对低照度室内视频烟雾图像进行研究,通过检测并识别图像中的烟雾目标来实现该条件下的火灾预警。在低照度室内环境下,利用红外摄像头获取的视频图像一般都存在脉冲噪声较多、视觉信息不够丰富等缺点,为了去除图像噪声和凸显图像中的目标,本文通过对自适应中值滤波算法进行改进,提出了一种低照度视频图像的去噪新算法。为了降低低照度室内光线变化给目标检测带来的影响和获得噪声更少的烟雾目标图像,通过对混合高斯模型法中的学习率进行改进,以此实现烟雾目标的检测。检测出烟雾目标后,采用先开运算再膨胀运算的方式来填充目标内部空洞和消除目标图像中的小块无意义区域。获得精确完整的目标区域后,为了区分烟雾目标,分别对目标的面积扩散特征、运动方向特征和纹理特征进行研究。针对面积扩散特征,采用计算一段时间内的绝对面积变化率的平均值来表征烟雾目标的面积扩散特征;针对运动方向特征,采用计算目标轮廓最小外围矩形中心点主运动方向比例来表征烟雾目标的扩散方向特征;针对纹理特征,采用计算烟雾目标原图矩形区域的对比度和二阶矩来表征烟雾目标的纹理特征。为了提高烟雾目标的识别率,采用基于径向基核函数的支持向量机进行烟雾目标特征的融合与判定。为了验证算法的识别能力,分别对算法进行了场景、监控距离、室内光照度、烟雾浓度以及存在干扰物条件下的测试,测试结果表明,本文算法在大多数情况下可以有效区分烟雾和干扰物,烟雾检测率高,具备较强的抗干扰能力和较好的场景适应能力。
[Abstract]:Smoke detection based on video image is a new fire detection technology in the field of fire protection. For video smoke detection, most of the research at home and abroad is in good light environment. Although the research on low illumination condition is still in its infancy, it has a very important guiding significance for fire prevention. The fire under low illumination not only causes more casualties, but also causes serious damage to people's property. The main work of this paper is to study the low-illuminance indoor video smoke images and to detect and recognize the smoke targets in the images to realize the fire warning under this condition. In the low illuminance indoor environment, the video images obtained by infrared camera generally have many disadvantages, such as pulse noise and insufficient visual information. In order to remove the image noise and highlight the target in the image, By improving the adaptive median filtering algorithm, a new de-noising algorithm for low-illuminance video images is proposed in this paper. In order to reduce the influence of low illuminance indoor light change on target detection and obtain smog target images with less noise, the learning rate of mixed Gao Si model method is improved to achieve smoke target detection. After the smoke target is detected, the hole inside the target is filled and the meaningless area in the target image is eliminated by the way of first opening operation and then expanding operation. After obtaining the accurate and complete target area, in order to distinguish the smog target, the area diffusion feature, motion direction feature and texture feature of the target are studied respectively. According to the characteristics of area diffusion, the average value of absolute area change rate over a period of time is used to characterize the area diffusion characteristics of smog targets, and the characteristics of moving direction are analyzed. The diffusive direction characteristics of smog target are represented by calculating the proportion of the main moving direction of the minimum peripheral rectangular center point of the target contour, and the texture features are analyzed. The contrast and second order moments of the rectangular region of the original map of the smoke target are calculated to characterize the texture feature of the smoke target. In order to improve the recognition rate of smoke target, support vector machine based on radial basis function kernel function is used to fuse and judge the feature of smoke target. In order to verify the recognition ability of the algorithm, the scene, monitoring distance, indoor illumination, smoke concentration and the presence of interference were tested respectively. The test results show that, In most cases, the proposed algorithm can effectively distinguish smoke from interference, with high smoke detection rate, strong anti-jamming ability and better scene adaptability.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】