【摘要】：煤礦井下視頻作業(yè)環(huán)境復(fù)雜、噪聲大、光照不均、存在遮擋以及僅依靠人工職守智能化低、誤檢率高等問(wèn)題。有必要研究圖像增強(qiáng)、提取特征點(diǎn)快速查詢(xún)配準(zhǔn)、實(shí)時(shí)動(dòng)態(tài)目標(biāo)檢測(cè)、遺留物檢測(cè)以及目標(biāo)跟蹤等技術(shù)。并通過(guò)理解分析圖像畫(huà)面出現(xiàn)的違規(guī)行為、可疑目標(biāo)和潛在危險(xiǎn),以快速合理的方式發(fā)出聯(lián)動(dòng)報(bào)警,同時(shí)為事故后期分析提供第一手資料。本文在對(duì)煤礦圖像處理相關(guān)技術(shù)研究的基礎(chǔ)上,針對(duì)煤礦安全生產(chǎn)等方面問(wèn)題,研究煤礦井下動(dòng)態(tài)目標(biāo)視頻監(jiān)測(cè)圖像處理的相關(guān)算法。針對(duì)礦井圖像受噪聲大導(dǎo)致畫(huà)面不清等問(wèn)題,在分析研究現(xiàn)有圖像增強(qiáng)相關(guān)技術(shù)的基礎(chǔ)上,提出一種基于模糊熵判別準(zhǔn)則合理提取局部模糊分形維數(shù)(LFFD)的相似度增強(qiáng)算法,該算法通過(guò)模糊熵判別合理LFFD融合相似性測(cè)度來(lái)調(diào)整圖像對(duì)比度,并考慮增強(qiáng)過(guò)程中的多參數(shù)性在相似度測(cè)量理論上的應(yīng)用。通過(guò)陜西韓城象山煤礦井下錨網(wǎng)支護(hù)圖像實(shí)驗(yàn)結(jié)果表明,該算法能較好抑制噪聲提高圖像對(duì)比度。針對(duì)如何獲得恰當(dāng)?shù)木略O(shè)備、支護(hù)等關(guān)鍵部件的圖像特征及細(xì)節(jié)紋理,考慮井下受到拍攝環(huán)境噪聲等因素對(duì)特征邊緣的影響,進(jìn)而提出一種基于小波分解的Canny邊緣檢測(cè)算法。該算法引入小波變換提取灰度圖像的高低頻分量,以此來(lái)獲得更多的邊緣信息完善特征輪廓,并為特征點(diǎn)云的精確收集起到關(guān)鍵作用。針對(duì)煤礦井下視頻圖像受粉塵、光照等干擾導(dǎo)致監(jiān)控圖像質(zhì)量下降以及煤礦視頻監(jiān)控系統(tǒng)采集點(diǎn)多,歷史留存數(shù)據(jù)量大不利于后續(xù)查找特征圖像等問(wèn)題。本文提出一種基于相關(guān)法的歐式距離配準(zhǔn)算法,該算法通過(guò)利用不同特征點(diǎn)自身信息,在Harris算法基礎(chǔ)上分別對(duì)灰度信息使用梯度相關(guān)法,對(duì)SIFT算法描述子信息使用描述子相關(guān)法,并結(jié)合特征點(diǎn)間的歐式距離關(guān)系來(lái)精確匹配,煤礦井下圖像匹配實(shí)驗(yàn)結(jié)果表明,本文算法降低了誤匹配的點(diǎn)數(shù)。針對(duì)煤礦視頻照度不均、噪聲大等環(huán)境極易丟失目標(biāo)以及煤區(qū)安全生產(chǎn)對(duì)排查前景目標(biāo)精度要求高等問(wèn)題。研究基于碼書(shū)模型(CBM)的運(yùn)動(dòng)目標(biāo)檢測(cè)算法,針對(duì)當(dāng)目標(biāo)的運(yùn)動(dòng)信息不足時(shí),CBM可能會(huì)出現(xiàn)誤檢或局部漏測(cè)等問(wèn)題。通過(guò)聯(lián)合目標(biāo)的空間整體性,提出一種基于CBM的目標(biāo)空間整體性背景更新算法,該算法通過(guò)對(duì)運(yùn)動(dòng)目標(biāo)空間信息變化分析,尋找前景中潛在的背景,并聯(lián)合像素時(shí)域統(tǒng)計(jì)進(jìn)行背景模型更新。實(shí)驗(yàn)結(jié)果表明,該算法可以快速適應(yīng)背景變化,在處理緩慢移動(dòng)目標(biāo)和只有局部運(yùn)動(dòng)目標(biāo)時(shí)能減少由于運(yùn)動(dòng)信息不足所造成的誤判,同時(shí)保證目標(biāo)檢測(cè)的完整性。針對(duì)目標(biāo)檢測(cè)時(shí)受陰影干擾等問(wèn)題。進(jìn)而提出一種基于HSV空間的碼字分量平均算法,該算法通過(guò)構(gòu)建碼字加權(quán)平均背景模型,并將RGB空間轉(zhuǎn)換成HSV空間達(dá)到更新背景去除陰影的效果。實(shí)驗(yàn)結(jié)果表明,算法對(duì)去除陰影有較強(qiáng)的魯棒性。針對(duì)煤礦膠帶運(yùn)輸機(jī)遺留物可能對(duì)膠帶以及滾筒等設(shè)備造成的損傷問(wèn)題。研究發(fā)現(xiàn)以多層背景模型為基礎(chǔ)的算法,通過(guò)控制不同模型的更新速度,比較模型之間的差異來(lái)判斷遺留物。這類(lèi)算法檢測(cè)速度較慢,對(duì)“鬼影”檢測(cè)存在誤檢。提出一種基于歷史像素穩(wěn)定度的遺留物檢測(cè)算法,該算法在運(yùn)動(dòng)目標(biāo)檢測(cè)的基礎(chǔ)上,對(duì)不屬于背景碼書(shū)模型的像素點(diǎn)記錄其之前若干幀像素的信息,構(gòu)成歷史像素集,并通過(guò)統(tǒng)計(jì)當(dāng)前像素與歷史像素集的匹配程度來(lái)判決該像素點(diǎn)是否穩(wěn)定,進(jìn)而判斷是否存在遺留物。并通過(guò)煤礦膠帶輸送機(jī)的視頻對(duì)該方法進(jìn)行了驗(yàn)證。針對(duì)煤礦動(dòng)態(tài)目標(biāo)的復(fù)雜運(yùn)動(dòng)、光照變化以及遮擋等因素對(duì)目標(biāo)跟蹤性能的影響。而現(xiàn)有基于多特征融合的跟蹤算法在復(fù)雜環(huán)境下跟蹤準(zhǔn)確度不高,且大部分采用單一判定方式來(lái)實(shí)現(xiàn)多特征融合的問(wèn)題。提出一種基于多特征判定準(zhǔn)則的目標(biāo)跟蹤融合算法,該算法首先引入局部背景信息加強(qiáng)對(duì)目標(biāo)的描述,其次在多特征融合過(guò)程中利用多種判定準(zhǔn)則自適應(yīng)計(jì)算特征權(quán)值,然后在Mean Shift框架下,結(jié)合Kalman濾波完成對(duì)目標(biāo)的跟蹤。陜西張家峁煤礦井下視頻實(shí)驗(yàn)結(jié)果表明,該算法比單種判定融合有更好的穩(wěn)定性和魯棒性,能有效地提高復(fù)雜環(huán)境下跟蹤準(zhǔn)確性。
[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，

本文編號(hào)：2272264