煤矿智能视频监控系统关键技术的研究

发布时间：2018-06-13 18:13

本文选题：煤矿智能视频监控系统 + 去雾除尘和同步去噪　；参考：《中国矿业大学》2013年博士论文

【摘要】：目前我国大多数的煤矿视频监控系统还主要停留在人工监控阶段，智能化煤矿视频监控系统是发展的必然趋势。它可以自动采集获得视频监控图像序列，进行实时运动目标检测、识别和跟踪，通过理解分析图像画面主动发现违规行为、可疑目标和潜在危险，以快速合理的方式发出警报，指导启动相应的联动控制措施。煤矿智能视频监控系统的实现，需要综合运用图像处理、机器学习和计算机视觉等领域中的多项技术，本文对其中的四类关键技术进行研究，具体工作包括：为了对伴有随机噪声的煤矿雾尘图像进行清晰化处理，提出一种基于DCPBF的去雾除尘和同步去噪算法。推导建立煤矿雾尘降质图像退化模型；设计基于暗原色先验知识的环境光、粗略透射率估计方法与步骤；采用联合双边滤波快速获得精细透射率图；依据图像退化模型构建正则化目标函数，求取转换图像并进行高斯双边滤波，获得去雾除尘图像且同步实现噪声的有效去除。针对相对静止的煤矿视频监控环境背景，采用背景减除法进行运动目标检测。提出基于聚类技术的自适应背景建模与更新方法，利用改进的FCM算法对像素灰度取值进行聚类，自适应选取不同个数的聚类构建各像素背景模型，随场景变化进行聚类修改、添加和删除完成背景更新。联合背景差分信息、三帧差分信息和空间邻域信息进行前景检测，通过改进的OTSU方法自动设置差分阈值。提出结合像素亮度和纹理特征的运动阴影检测方法，依据在阴影覆盖前后的灰度图像中，像素具有亮度值相关性和纹理特征值不变性，实现运动阴影的检测与去除。将单目标跟踪看作为目标和背景的在线分类问题，选用线性SVM作为分类工具，提出一种添加样本约简机制的FLSVMIL方法实现分类器在线更新，并提出基于FLSVMIL的单目标跟踪算法。由于可能受到无效历史信息的干扰，并且难以处理样本集非线性可分的问题，提出基于LSVMSE的单目标跟踪算法，采用集成分类器进行运动目标跟踪。根据煤矿智能视频监控系统中多目标跟踪的任务需求，提出基于UKF-MHT的多目标跟踪算法。设计算法的基本框架，确定关键步骤的处理方法，其中包括跟踪门设置、目标预测值与观测值的数据匹配、航迹评价与删除、航迹聚类和m-best假设的产生以及目标状态的预测更新。在自适应跟踪修正阶段，针对由目标短暂丢失、粘连和分裂可能引起的三类跟踪错误，，设计具体的判别策略和修正方法。
[Abstract]:At present, most coal mine video surveillance systems in China are still mainly in the stage of manual monitoring, and intelligent coal mine video surveillance system is an inevitable trend of development. It can automatically capture and obtain video surveillance image sequences, detect, identify and track moving objects in real time, actively discover irregularities, suspicious targets and potential dangers through understanding and analyzing images, and issue warnings in a rapid and reasonable manner. Guide to initiate the corresponding linkage control measures. In order to realize the intelligent video surveillance system in coal mine, it is necessary to comprehensively apply many techniques in the fields of image processing, machine learning and computer vision. In this paper, four kinds of key technologies are studied. The main work includes: in order to clear the dust image with random noise, a DCPBF based de-fogging and simultaneous de-noising algorithm is proposed. The degradation model of degraded image of coal mine fog dust is established, the environmental light and rough transmittance estimation method and steps based on the prior knowledge of dark primary color are designed, and the fine transmittance map is obtained quickly by combined bilateral filtering. According to the image degradation model, the regularization objective function is constructed, the converted image is obtained and Gao Si bilateral filtering is carried out, and the de-fogging and dedusting image is obtained, and the effective noise removal is realized synchronously. The background subtraction method is used to detect moving targets against the background of relatively static coal mine video surveillance environment. An adaptive background modeling and updating method based on clustering technology is proposed. The improved FCM algorithm is used to cluster the pixel grayscale, and the background model of each pixel is constructed by selecting different numbers of clustering adaptively, and the background model is modified with the change of scene. Add and delete complete background updates. The background differential information, the three frame differential information and the spatial neighborhood information are combined to detect the foreground, and the differential threshold is automatically set by the improved Otsu method. A method of moving shadow detection combined with pixel luminance and texture features is proposed. According to the correlation of luminance values and the invariance of texture feature values in gray images before and after shadow coverage the detection and removal of moving shadows are realized. Considering single target tracking as an online classification problem of target and background, linear SVM is used as a classification tool. A FLSVMIL method with sample reduction mechanism is proposed to update the classifier online, and a single target tracking algorithm based on FLSVMIL is proposed. Because of the disturbance of invalid historical information and the difficulty of dealing with the nonlinear separable problem of the sample set, a single target tracking algorithm based on LSVMSE is proposed, and an integrated classifier is used for moving target tracking. According to the task requirement of multi-target tracking in intelligent video surveillance system of coal mine, a multi-target tracking algorithm based on UKF-MHT is proposed. Design the basic framework of the algorithm, determine the key steps of the processing methods, including tracking gate setting, target prediction and observation data matching, track evaluation and deletion, Track clustering, m-best hypothesis generation and target state prediction update. In the stage of adaptive tracking correction, a specific discriminant strategy and correction method are designed for three kinds of tracking errors which may be caused by the transient loss, adhesion and splitting of the target.
【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TD76;TP273

【参考文献】