煤矿井下动态目标视频监测图像处理研究
发布时间：2018-10-15 10:30
【摘要】：煤矿井下视频作业环境复杂、噪声大、光照不均、存在遮挡以及仅依靠人工职守智能化低、误检率高等问题。有必要研究图像增强、提取特征点快速查询配准、实时动态目标检测、遗留物检测以及目标跟踪等技术。并通过理解分析图像画面出现的违规行为、可疑目标和潜在危险,以快速合理的方式发出联动报警,同时为事故后期分析提供第一手资料。本文在对煤矿图像处理相关技术研究的基础上,针对煤矿安全生产等方面问题,研究煤矿井下动态目标视频监测图像处理的相关算法。针对矿井图像受噪声大导致画面不清等问题,在分析研究现有图像增强相关技术的基础上,提出一种基于模糊熵判别准则合理提取局部模糊分形维数(LFFD)的相似度增强算法,该算法通过模糊熵判别合理LFFD融合相似性测度来调整图像对比度,并考虑增强过程中的多参数性在相似度测量理论上的应用。通过陕西韩城象山煤矿井下锚网支护图像实验结果表明,该算法能较好抑制噪声提高图像对比度。针对如何获得恰当的井下设备、支护等关键部件的图像特征及细节纹理,考虑井下受到拍摄环境噪声等因素对特征边缘的影响,进而提出一种基于小波分解的Canny边缘检测算法。该算法引入小波变换提取灰度图像的高低频分量,以此来获得更多的边缘信息完善特征轮廓,并为特征点云的精确收集起到关键作用。针对煤矿井下视频图像受粉尘、光照等干扰导致监控图像质量下降以及煤矿视频监控系统采集点多,历史留存数据量大不利于后续查找特征图像等问题。本文提出一种基于相关法的欧式距离配准算法,该算法通过利用不同特征点自身信息,在Harris算法基础上分别对灰度信息使用梯度相关法,对SIFT算法描述子信息使用描述子相关法,并结合特征点间的欧式距离关系来精确匹配,煤矿井下图像匹配实验结果表明,本文算法降低了误匹配的点数。针对煤矿视频照度不均、噪声大等环境极易丢失目标以及煤区安全生产对排查前景目标精度要求高等问题。研究基于码书模型(CBM)的运动目标检测算法,针对当目标的运动信息不足时,CBM可能会出现误检或局部漏测等问题。通过联合目标的空间整体性,提出一种基于CBM的目标空间整体性背景更新算法,该算法通过对运动目标空间信息变化分析,寻找前景中潜在的背景,并联合像素时域统计进行背景模型更新。实验结果表明,该算法可以快速适应背景变化,在处理缓慢移动目标和只有局部运动目标时能减少由于运动信息不足所造成的误判,同时保证目标检测的完整性。针对目标检测时受阴影干扰等问题。进而提出一种基于HSV空间的码字分量平均算法,该算法通过构建码字加权平均背景模型,并将RGB空间转换成HSV空间达到更新背景去除阴影的效果。实验结果表明,算法对去除阴影有较强的鲁棒性。针对煤矿胶带运输机遗留物可能对胶带以及滚筒等设备造成的损伤问题。研究发现以多层背景模型为基础的算法,通过控制不同模型的更新速度,比较模型之间的差异来判断遗留物。这类算法检测速度较慢,对“鬼影”检测存在误检。提出一种基于历史像素稳定度的遗留物检测算法,该算法在运动目标检测的基础上,对不属于背景码书模型的像素点记录其之前若干帧像素的信息,构成历史像素集,并通过统计当前像素与历史像素集的匹配程度来判决该像素点是否稳定,进而判断是否存在遗留物。并通过煤矿胶带输送机的视频对该方法进行了验证。针对煤矿动态目标的复杂运动、光照变化以及遮挡等因素对目标跟踪性能的影响。而现有基于多特征融合的跟踪算法在复杂环境下跟踪准确度不高,且大部分采用单一判定方式来实现多特征融合的问题。提出一种基于多特征判定准则的目标跟踪融合算法,该算法首先引入局部背景信息加强对目标的描述,其次在多特征融合过程中利用多种判定准则自适应计算特征权值,然后在Mean Shift框架下,结合Kalman滤波完成对目标的跟踪。陕西张家峁煤矿井下视频实验结果表明,该算法比单种判定融合有更好的稳定性和鲁棒性,能有效地提高复杂环境下跟踪准确性。
[Abstract]:The underground video operation environment of the coal mine is complicated, the noise is large, the illumination is not uniform, the shielding is existed, and the problem of high false detection rate and the like is realized only by virtue of the intelligent low-intelligence and the false detection rate. It is necessary to study image enhancement, extract feature point fast query matching, real-time dynamic target detection, object detection and target tracking. And by understanding the violation, suspicious object and potential danger appearing in the image picture, the linkage alarm is sent out in a fast and reasonable way, and the first-hand data is provided for the post-accident analysis. Based on the research of coal mine image processing technology, this paper studies the related algorithms of image processing of dynamic target video in coal mine, aiming at the problems of coal mine safety production and so on. On the basis of analyzing and studying the related art of image enhancement, a similarity enhancement algorithm based on fuzzy entropy discrimination criterion to extract local fuzzy fractal dimension number (LFFD) is proposed. The algorithm adjusts the image contrast by using the fuzzy entropy discrimination reasonable LFFD fusion similarity measure, and considers the application of multi-parameter in the enhancement process on the similarity measurement theory. The experimental results show that this algorithm can restrain the noise and improve the image contrast. Aiming at how to obtain the image features and detail textures of key parts such as well equipment, support and so on, considering the influence of factors such as ambient noise and other factors on the characteristic edge, a Canny edge detection algorithm based on wavelet decomposition is proposed. The algorithm introduces the wavelet transform to extract the high low-frequency component of the gray-scale image, so as to obtain more edge information and perfect the characteristic contour, and plays a key role in the accurate collection of the feature point cloud. aiming at the problems that the monitoring image quality is degraded due to the interference of dust, light and the like in the underground video image of the coal mine, the collection points of the coal mine video monitoring system are much, the history retention data amount is large, and the subsequent searching feature images and the like are difficult to follow. This paper presents a kind of Euclidean distance matching algorithm based on correlation method, which uses the information of different feature points to use gradient correlation method on the gray information on the basis of Harris algorithm, and describes the sub-information using description sub-information using the SIFT algorithm. According to the European distance relation between feature points, the experiment results show that the algorithm reduces the number of mis-matching. aiming at the problems such as uneven illumination of the coal mine, high noise and the like, and the high requirement of the safety production of the coal area on the detection prospect target precision. In this paper, a motion target detection algorithm based on code book model (CBM) is studied. When the motion information of the target is insufficient, the CBM may have problems such as false detection or local leakage. Based on the spatial integrity of the joint target, a new algorithm for updating the global background of target space based on CBM is proposed, which is based on the analysis of the spatial information of the moving object, finds the potential background in the foreground, and updates the background model based on the time-domain statistics of the pixels. The experimental results show that the algorithm can rapidly adapt to the background changes, and can reduce the misjudgment caused by insufficient motion information when dealing with slow moving targets and only local moving targets, while ensuring the integrity of target detection. Aiming at the problems such as shadow interference and the like in the detection of the target. Furthermore, a codeword component averaging algorithm based on HSV space is proposed. The algorithm is used to construct codeword weighted average background model and convert RGB space into HSV space to achieve the effect of updating background removal shadow. The experimental results show that the algorithm has strong robustness to the removal of shadows. Aiming at the damage caused by adhesive tape and roller and other equipment in coal mine belt conveyor. The research findings are based on multi-layer background model. By controlling the updating speed of different models, the differences between models are compared. The detection speed of this class algorithm is slow, and there is a false check for the 鈥済host鈥，
本文编号：2272264
[Abstract]:The underground video operation environment of the coal mine is complicated, the noise is large, the illumination is not uniform, the shielding is existed, and the problem of high false detection rate and the like is realized only by virtue of the intelligent low-intelligence and the false detection rate. It is necessary to study image enhancement, extract feature point fast query matching, real-time dynamic target detection, object detection and target tracking. And by understanding the violation, suspicious object and potential danger appearing in the image picture, the linkage alarm is sent out in a fast and reasonable way, and the first-hand data is provided for the post-accident analysis. Based on the research of coal mine image processing technology, this paper studies the related algorithms of image processing of dynamic target video in coal mine, aiming at the problems of coal mine safety production and so on. On the basis of analyzing and studying the related art of image enhancement, a similarity enhancement algorithm based on fuzzy entropy discrimination criterion to extract local fuzzy fractal dimension number (LFFD) is proposed. The algorithm adjusts the image contrast by using the fuzzy entropy discrimination reasonable LFFD fusion similarity measure, and considers the application of multi-parameter in the enhancement process on the similarity measurement theory. The experimental results show that this algorithm can restrain the noise and improve the image contrast. Aiming at how to obtain the image features and detail textures of key parts such as well equipment, support and so on, considering the influence of factors such as ambient noise and other factors on the characteristic edge, a Canny edge detection algorithm based on wavelet decomposition is proposed. The algorithm introduces the wavelet transform to extract the high low-frequency component of the gray-scale image, so as to obtain more edge information and perfect the characteristic contour, and plays a key role in the accurate collection of the feature point cloud. aiming at the problems that the monitoring image quality is degraded due to the interference of dust, light and the like in the underground video image of the coal mine, the collection points of the coal mine video monitoring system are much, the history retention data amount is large, and the subsequent searching feature images and the like are difficult to follow. This paper presents a kind of Euclidean distance matching algorithm based on correlation method, which uses the information of different feature points to use gradient correlation method on the gray information on the basis of Harris algorithm, and describes the sub-information using description sub-information using the SIFT algorithm. According to the European distance relation between feature points, the experiment results show that the algorithm reduces the number of mis-matching. aiming at the problems such as uneven illumination of the coal mine, high noise and the like, and the high requirement of the safety production of the coal area on the detection prospect target precision. In this paper, a motion target detection algorithm based on code book model (CBM) is studied. When the motion information of the target is insufficient, the CBM may have problems such as false detection or local leakage. Based on the spatial integrity of the joint target, a new algorithm for updating the global background of target space based on CBM is proposed, which is based on the analysis of the spatial information of the moving object, finds the potential background in the foreground, and updates the background model based on the time-domain statistics of the pixels. The experimental results show that the algorithm can rapidly adapt to the background changes, and can reduce the misjudgment caused by insufficient motion information when dealing with slow moving targets and only local moving targets, while ensuring the integrity of target detection. Aiming at the problems such as shadow interference and the like in the detection of the target. Furthermore, a codeword component averaging algorithm based on HSV space is proposed. The algorithm is used to construct codeword weighted average background model and convert RGB space into HSV space to achieve the effect of updating background removal shadow. The experimental results show that the algorithm has strong robustness to the removal of shadows. Aiming at the damage caused by adhesive tape and roller and other equipment in coal mine belt conveyor. The research findings are based on multi-layer background model. By controlling the updating speed of different models, the differences between models are compared. The detection speed of this class algorithm is slow, and there is a false check for the 鈥済host鈥，
本文编号：2272264
本文链接：https://www.wllwen.com/kejilunwen/anquangongcheng/2272264.html
