本文選題：線激光測(cè)量　切入點(diǎn)：特征提取　出處：《廣東工業(yè)大學(xué)》2015年碩士論文　論文類型：學(xué)位論文

【摘要】：工業(yè)檢測(cè)中,隨著被測(cè)工件越來(lái)越復(fù)雜和人們要求測(cè)量精度越來(lái)越高,傳統(tǒng)機(jī)器視覺(jué)測(cè)量因其二維局限性已滿足不了精密檢測(cè)的要求。與此同時(shí),激光測(cè)量技術(shù)因?yàn)轸敯粜詮?qiáng),精確度高,越來(lái)越受到人們的歡迎。本文針對(duì)實(shí)際項(xiàng)目需求,設(shè)計(jì)和搭建線激光測(cè)量系統(tǒng)。主要闡述硬件結(jié)構(gòu)的設(shè)計(jì)和軟件框架的搭建,并對(duì)程序設(shè)計(jì)流程作了分析。由于線激光設(shè)備具有掃描盲區(qū),面對(duì)較復(fù)雜的被測(cè)工件時(shí)無(wú)法一次性獲得整個(gè)工件的三維數(shù)據(jù),有時(shí)需要多次在不同視角下進(jìn)行掃描,然后對(duì)三維數(shù)據(jù)進(jìn)行拼接以獲得整個(gè)被測(cè)工件模型。鑒于此,本文同時(shí)對(duì)三維點(diǎn)云拼接算法進(jìn)行了研究。在點(diǎn)云特征提取中,針對(duì)線激光掃描的點(diǎn)云數(shù)據(jù)的結(jié)構(gòu)特點(diǎn),提出將三維點(diǎn)云模型映射到二維圖像中,并修改圖像處理中的Canny算法,對(duì)點(diǎn)云模型進(jìn)行特征潛在區(qū)域估算,然后基于曲率信息提取特征點(diǎn)。與傳統(tǒng)的三維點(diǎn)云特征提取算法比較,該方法具有更高的執(zhí)行效率。在點(diǎn)云特征描述中,高維特征算子雖然可以直接解決多視圖三維點(diǎn)云數(shù)據(jù)的配準(zhǔn)問(wèn)題,但它的計(jì)算較復(fù)雜,對(duì)大數(shù)量級(jí)的點(diǎn)云模型來(lái)說(shuō),匹配效率將大大降低。本文出于運(yùn)行效率的考慮,提出基于鄰域曲率信息的描述方法,用低維描述算子代替高維描述算子,減少計(jì)算時(shí)間。在點(diǎn)云模型間對(duì)應(yīng)點(diǎn)匹配中,根據(jù)特征描述算子構(gòu)造評(píng)價(jià)函數(shù),為每個(gè)特征點(diǎn)尋找潛在對(duì)應(yīng)點(diǎn)。由于低維特征描述的信息豐富性不強(qiáng),每個(gè)特征點(diǎn)往往匹配到多個(gè)對(duì)應(yīng)點(diǎn),于是需要作進(jìn)一步篩選。本文基于剛性變換原則,提出區(qū)域投票制度的方式實(shí)現(xiàn)對(duì)應(yīng)點(diǎn)的精匹配。本文最后對(duì)拼接算法進(jìn)行了實(shí)驗(yàn)和對(duì)設(shè)計(jì)的線激光測(cè)量機(jī)進(jìn)行了平整度檢測(cè)應(yīng)用測(cè)試。首先,測(cè)量機(jī)對(duì)實(shí)際物體多次掃描得到不同視角下的點(diǎn)云數(shù)據(jù)模型,通過(guò)本文提出的特征提取算子、特征描述算子和對(duì)應(yīng)點(diǎn)匹配算法對(duì)兩點(diǎn)云模型進(jìn)行特征分析。然后,基于迭代最近點(diǎn)算法對(duì)兩點(diǎn)云模型進(jìn)行拼接,驗(yàn)證了拼接算法的正確性和有效性。在實(shí)際應(yīng)用測(cè)試中,本文對(duì)線激光測(cè)量機(jī)的標(biāo)定做了實(shí)驗(yàn)分析,并利用線激光測(cè)量機(jī)的平整度檢測(cè)功能,對(duì)幾種平整度要求較嚴(yán)格的工件進(jìn)行實(shí)際測(cè)量,給出測(cè)量結(jié)果和工件的3D掃描圖形,證明測(cè)量機(jī)功能的有效性。
[Abstract]:In industrial detection, with the increasing complexity of the workpiece under test and the increasing requirement of measuring precision, the traditional machine vision measurement can not meet the requirement of precision detection because of its two-dimensional limitation. At the same time, the laser measurement technology has strong robustness. This paper designs and builds the line laser measurement system according to the actual project demand, mainly expounds the design of the hardware structure and the construction of the software frame. The program design flow is analyzed. Because the line laser equipment has scanning blind area, the 3D data of the whole workpiece can not be obtained at one time in the face of the more complex workpiece being tested, and sometimes it is necessary to scan the whole workpiece in different angle of view several times. Then the 3D data are spliced to obtain the whole workpiece model. In view of this, the algorithm of 3D point cloud mosaic is also studied in this paper. In the feature extraction of point cloud, the structural characteristics of the point cloud data scanned by line laser are analyzed. The 3D point cloud model is mapped to two-dimensional image, and the Canny algorithm in image processing is modified to estimate the potential feature region of the point cloud model. Then the feature points are extracted based on curvature information. Compared with the traditional 3D point cloud feature extraction algorithm, this method has higher execution efficiency. Although the high dimensional feature operator can directly solve the registration problem of multi-view 3D point cloud data, its calculation is more complicated, and the matching efficiency will be greatly reduced for the point cloud model with large order of magnitude. In this paper, a description method based on neighborhood curvature information is proposed, in which the low dimensional description operator is used instead of the high dimensional description operator to reduce the computing time. In the point cloud model corresponding point matching, the evaluation function is constructed according to the characteristic description operator. Finding potential correspondence points for each feature point. Because of the low dimensional feature description is not rich in information, each feature point often matches to more than one corresponding point, so we need to make further screening. This paper based on rigid transformation principle, In this paper, the method of regional voting system is proposed to realize the precise matching of corresponding points. Finally, experiments are carried out on the splicing algorithm and the flatness detection and application test of the designed line laser measuring machine are carried out. The point cloud data model with different angle of view is obtained by scanning the actual object several times. The feature extraction operator, feature description operator and corresponding point matching algorithm are used to analyze the feature of the two-point cloud model. Two point cloud model is stitched based on iterative nearest point algorithm, which verifies the correctness and validity of the stitching algorithm. In the practical application test, the calibration of the line laser measuring machine is analyzed experimentally. By using the flatness detection function of the line laser measuring machine, the actual measurement of several workpieces with strict flatness requirements is carried out, and the measuring results and 3D scanning figure of the workpiece are given, which proves the validity of the function of the measuring machine.
【學(xué)位授予單位】：廣東工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：TN249;TP391.41
，

本文編號(hào)：1640869