基於線掃描式攝影機之高解析度雙目立體視覺量測技術

High resolution measurement utilizing a line scan stereo vision technique

陳柏呈(Po-Cheng Chen)

指導教授 ： 章明

中原大學/工學院/機械工程研究所/博士(2017年)

https://doi.org/10.6840/cycu201700928

摘要

本研究以傳統雙目立體視覺為基礎，結合兩組線掃描式攝影機與高速線型移動平台同步連續取像，發展出一套基於線掃描式攝影機之創新式高解析度雙目立體視覺量測技術，應用於待測物表面缺陷檢測與三維輪廓重建。
兩組相機之視角自相異位置交會並聚焦於線性移動載台平面，形成一對焦基準線，其中一組相機視角需與平台移動方向垂直;循兩相機掃描取像方向可延伸為一虛擬基準平面。以垂直視角取得之二維影像為基準，影像上任意點座標(x,y)，分別以線掃描條數與單一掃瞄線之像素間距決定。待測物之三維輪廓與高度重建，取決於搜尋兩張影像中相同物理對應點之準確性，本研究採用以數位影像相關法進行匹配運算，以正規化交叉匹配相關函數作為演算核心。利用空間與數位影像之映射關係，建立影像畸變校正之座標轉換方程式，完成簡單且有效率的校正程序。根據兩台攝影機個別取得待測點影像之掃描線數量差異後，可計算出高度h（x，y）；其解析度取決於單條線掃描影像所擷取之樣本位移量而定。藉由步階塊規組合、球柵陣列封裝(BGA)底部陣列式錫球引腳高度與共面度量測確認系統精度;最後採用我國十元硬幣進行三維表面形貌重建，硬幣上的日期和浮雕圖案皆可明顯辨識。
為驗證本研究成果具備線上檢測之能力，亦建立一套覆晶封裝製程之封膠厚度（BLT）與晶片傾斜度之自動量測系統。使用前述成像和測量原理，首先採用一系列影像前處理，進行晶片與相鄰定位點之邊緣強化;分別透過霍氏轉換直線與圓形偵測找出晶片之角點與基板定位點座標，求出四角點之相對高度值，進而求得封膠厚度與晶片傾斜度。實驗使用之覆晶基板樣品尺寸為15×15mm2，自基板至晶片頂端之高度約為800~900μm，其中晶片厚度已知為800μm，得封膠厚度為76.4 μm，晶片傾斜度為0.006度，顯示本系統之軸向量測精度已可達次微米等級，未來有機會實際應用於產線上。

關鍵字

線掃描式攝影機 ； 雙目立體視覺量測技術 ； 三維形貌量測 ； 數位影像相關法 ； 封膠厚度

並列摘要

In this study, a high-resolution surface topography measurement system was developed using two line scan cameras for the full-field optical inspection of surface geometry and defects. The two cameras capture and record multiple line images of a test object simultaneously based on stereoscopic vision technique. Reconstruction algorithms were developed to reconstruct visual depth or 3D information of test objects. System performance was evaluated using different test specimens. Compared to area cameras of the same pixel resolution, the use of line scan cameras in the inspection system offers the advantages of processing efficiency, high sensitivity, and better resolution.
In the measurement, one camera vertically scans the object while the second camera captures the image of the same object from an angle. A datum line along the intersection of the two scan directions is assigned forming a datum surface with the scanning axis. Based on this surface, a 2D coordinate of an arbitrary point (x, y) from the image captured by the vertical camera is expressed in terms of the number scan lines and pitch between each scan line-image. Critical matching of the same physical points in the two images determines the 3D shape and height reconstruction of the object. To obtain the correspondence pairs from each line scan, a novel digital image correlation algorithm was developed. Calibration was established by mapping the horizontal coordinates of the object to the image coordinate in the cameras. The position difference in terms of pixel for each inspection point was identified by the algorithm. The height h(x, y) of the specimen was calculated from the difference in the number of scan lines between the two cameras. The horizontal resolution was set by the pixel spacing of the line scan cameras and the height resolution by the pitch of the line scan images. The height differences of three adjacent ladder-type gauge blocks were measured to establish the accuracy of the inspection system. The ball height and co-planarity of a ball grid array (BGA) were measured from 3D surface reconstruction, which identified overlooked defects from 2D inspection data. A Chinese ten-dollar coin specimen was reconstructed to reveal the date inscription and embossed profile on the coin. The 3D surface profiles of the test specimens were resolved with very good accuracy.
Aiming towards in-line measurement, the inspection system was automated for measurement of bond-line thickness (BLT) and die tilt in die attach chip. The same imaging and measurement principles were used. Initially, four points on the flip-chip substrate were marked in advance by the chip maker for reference. The positions of the four corners of the chip were determined from feature extraction on the specified region of interest using Hough transform, edge detection, image binarization, and image thinning from which four straight lines were obtained to delineate the edge of the chip surface. The 3D coordinates were determined from the intersection of two linear graphs on the chip and the center of the markings. The BLT and die tilt of the chip on a 800 mm thick substrate of cross section 15 x 15 mm2 were measured. The corresponding height and tilt were approximately 76.4 μm and 0.006, respectively. These measurements were verified experimentally, which shows that the inspection system is capable of providing sub-micrometer accuracy with high repeatability. Thus the automated measurement of the BLT on product line could be expected with the proposed technique.

並列關鍵字

Line scan cameras ； stereoscopic vision ； 3D surface profiles ； digital image correlation ； bond-line thickness

參考文獻

[30] K.-H. Lin, M. Chang, and T.-J. F. Chen, "Novel coordinate mapping algorithm for three-dimensional profile noncontact measurement," Optical Engineering, vol. 41, pp. 1615-1620, 2002.

[1] J. J. Park, K. Kwon, and N. Cho, "Development of a coordinate measuring machine (CMM) touch probe using a multi-axis force sensor," Measurement Science and Technology, vol. 17, pp. 2380-2386, 2006.

[2] S. H. R. Ali, "Probing System Characteristics in Coordinate Metrology," Measurement Science Review, vol. 10, pp. 120-129, 2010.

[3] I. Vrba, R. Palencar, M. Hadzistevic, B. Strbac, V. Spasic-Jokic, and J. Hodolic, "Different Approaches in Uncertainty Evaluation for Measurement of Complex Surfaces Using Coordinate Measuring Machine," Measurement Science Review, vol. 15, pp. 111-118, 2015.

[4] T. C. Niu and A. Li, "Exploring Single Molecules by Scanning Probe Microscopy: Porphyrin and Phthalocyanine," Journal of Physical Chemistry Letters, vol. 4, pp. 4095-4102, 2013.