本文選題：視覺測量　切入點：三維重構　出處：《武漢大學》2013年博士論文　論文類型：學位論文

【摘要】：隨著計算機技術的迅速發(fā)展和視覺傳感器成本的降低,視覺測量和重構技術依靠其非接觸、高精度、準實時化的優(yōu)點在制造和裝備業(yè)的產(chǎn)品品質管控(Quality Control,QC)中發(fā)揮著越來越重要的作用。近些年來,加工制造方法和工藝的改進,帶來了微(小)型產(chǎn)品(尺寸為2mm～100mm)的數(shù)量不斷增加,相對應的視覺測量任務也逐漸增多并且提出了一些新的要求,如檢測精度高、速度快、誤判率低等。本文針對微(小)型物體視覺測量與重構中的難點,以幾種微(小)型物體(電子連接器、醫(yī)學牙齒模型、醫(yī)用注射器、珠寶、小型非標零件)為研究對象,從二維和三維視覺測量兩個方面出發(fā),對微(小)型物體幾何特征提取和識別、精密外形快速重構、三維特征測量、模型簡化等具體問題進行了研究。論文的主要工作和創(chuàng)新點包括： 1.針對微(小)型電子產(chǎn)品的二維幾何和瑕疵的視覺測量及檢測需求,介紹了一種基于最小二乘和權重函數(shù)的幾何特征(直線、圓、橢圓)魯棒性提取算法,并以幾何特征作為骨架信息完成了對平面瑕疵的定位和識別。 2.針對微(小)型電子產(chǎn)品的表面特征識別需求,提出了一種較為通用的仿射不變特征識別算法。該算法綜合利用偽Zernike不變矩及SIFT匹配,能有效地解決發(fā)生旋轉、縮放、平移等特征的識別和分類。 3.針對紋理缺乏、形狀規(guī)則的全凸微(小)型物體快速三維重構需求,提出一種基于多視輪廓線的重構算法(Shape from closed contours,SFCC),能在極短的時間內(40sec以內)得到被測目標精密的mesh模型。該算法基于序列輪廓影像,通過初次切割和精密切割兩個步驟直接恢復面模型�；谠撍惴�,設計和裝配了相應的硬件設備,以鉆石和注射器針頭為實驗目標進行了測試,其中基于SFCC方法恢復的三維鉆石模型量測得到的線元素和角元素的精度分別可達到0.3mmm,0.03°,滿足鉆石切工等級監(jiān)控的需求。 4.針對復雜的不規(guī)則的微(小)型物體表面點云的拼接需求,提出一種基于旋轉平臺的快速拼接算法,該算法利用空間點云的柱面約束標定旋轉平臺,進而實現(xiàn)對任意角度點云的自動拼接,無需人工后處理和標志點約束,能很好的同各類三維掃描儀進行結合實現(xiàn)復雜物體的表面點云快速拼接,點云拼接精度同業(yè)類領先的拼接算法得到的結果相當(拼接誤差為0.3mm)。 5.基于點云數(shù)據(jù),推導了三維空間幾何特征(平面、球、圓柱體)的擬合公式,針對平面特征以平面度的提取進行了測試,針對球和圓柱體一款非標零件進行測試,完成了該零件上的球、圓柱特征的提取。 6.基于三角網(wǎng)數(shù)據(jù),提出了一種利用多方向斷面線約束的最優(yōu)格網(wǎng)簡化算法,該算法能高效的對密集三角網(wǎng)數(shù)據(jù)進行簡化,可分別輸出四邊形和三角形格網(wǎng)數(shù)據(jù),通過同原始數(shù)據(jù)的比對,簡化后的誤差控制在0.04mm以內。并以口腔學科中牙模應力性分析為測量需求,將本算法生成的牙模格網(wǎng)數(shù)據(jù)導入三維有限元分析軟件,成功的獲取到應力性分析數(shù)據(jù)。
[Abstract]:With the rapid development of computer technology and reduce the cost of the visual sensor, vision measurement and reconstruction technology relies on its non-contact, high accuracy, the advantages of quasi real time in the manufacturing and equipment industry product quality control (Quality, Control, QC) plays an increasingly important role. In recent years, the improved method and manufacturing process processing, bring the micro (small) products (the size of 2mm ~ 100mm) an increase in the number of visual measurement tasks corresponding to increase gradually and put forward some new requirements, such as high precision, fast speed, low false positive rate. In this paper, aiming at the micro (small) difficult objects in vision measurement and reconstruction, with several micro (small) object type (electronic connectors, medical dental model, medical syringe, jewelry, small non-standard parts) as the research object, from the 2D and 3D vision measurement based on two aspects of micro (small) type geometry Specific issues such as feature extraction and recognition, rapid contour reconstruction, 3D feature measurement and model simplification are studied.
1. for the micro (small) visual measurement and testing requirements of electronic products and two-dimensional geometric defects, introduces a kind of geometric features and least squares based on weight function (line, circle, ellipse) algorithm to extract the robustness, and geometric features as the skeleton information to complete the localization and identification of plane flaws.
2., aiming at the requirement of surface feature recognition for micro / small electronic products, a more general affine invariant feature recognition algorithm is proposed. The algorithm takes advantage of pseudo Zernike invariant moments and SIFT matching to effectively solve the recognition and classification of rotation, scaling and translation.
3. due to the lack of texture, the micro convex shape rules (small) object fast 3D reconstruction needs, proposed a multi view reconstruction algorithm based on contour (Shape from closed contours, SFCC), in a very short period of time (less than 40sec) measured mesh model precision. The algorithm is based on image through the initial contour sequence, cutting and precision cutting two step direct recovery models. Based on this algorithm, the design and assembly of the corresponding hardware equipment, diamond and syringe needle as the experimental targets were tested, the measured line elements and angular elements content of 3D diamond model based on SFCC method to restore the accuracy respectively. Can reach 0.3mmm, 0.03 degrees, to meet the diamond cut grade monitoring needs.
4. for the irregular complex micro (small) surface point cloud splicing demand, proposes a fast stitching algorithm based on rotating platform, the algorithm uses the cylindrical constraint space point cloud calibration rotary platform, so as to realize the automatic mosaic of arbitrary angle point cloud, without manual postprocessing and mark point constraint can, 3D scanner with all kinds of good combination of points on the surface of complex objects cloud fast mosaic, mosaic algorithm of point cloud splicing precision industry class leading results (error is 0.3mm).
5., based on the point cloud data, the fitting formula of three-dimensional spatial characteristics (plane, sphere and cylinder) is derived. The flatness extraction is tested for plane characteristics. A ball and cylinder part is tested for a non-standard part, and the extraction of spherical and cylindrical features on the part is completed.
6.鍩轟簬涓夎緗戞暟鎹，

本文編號：1589856