利用三度空間掃描資料擷取人體尺寸與體型

Collecting Body Sizes and Shapes from 3D Scanning Data

盧俊銘(Lu Jun-Ming)

指導教授 ： 王茂駿

國立清華大學/工學院/工業工程與工程管理學系/博士(2009年)

摘要

隨著客製化的需求與日俱增，對於人體尺寸與體型的詳盡了解與實際應用顯得更為重要。在過去的十年中，三度空間掃描技術已被廣泛使用於人體尺寸與體型的擷取，因此相關的研究也逐漸受到關注與期待。

為了有效地處理三度空間掃描資料，本研究提出一套自動化的特徵點辨識方法，以供後續的相關應用。經由分割掃描資料與初步的搜尋，輪廓分析、最小圍度測定、灰階分析、與人體等高線圖等四種演算法可輔助完成特徵點的辨識。透過189位受試者的掃描測試，並與傳統的人工觸診方法比較，本套方法被證實具有準確且精確的辨識能力。

根據特徵點的辨識結果，配合擷取尺寸的模擬方法，總計可以獲得104項的人體尺寸。隨後，本研究設計一個二階段的評估方法，透過準確度與精確度兩項指標評估利用三度空間掃描所擷取的人體尺寸。由263位受試者的掃描與量測結果發現，利用三度空間掃描所擷取的人體尺寸固然較傳統的人工量測方法精確，但兩者間存在的顯著差異說明了準確度的不足；為了控制受試者因素所造成的量測誤差，在第二階段的評估中，採用了服裝用人體模特兒進行掃描與手動量測，結果發現準確度與精確度皆有所提升並符合需求，顯見相關的改善是非常關鍵而有效的。此外，本研究亦探討掃描時的手部姿勢與受試者性別對於掃描量測結果的影響，根據實驗結果，採用掌心向後的手部姿勢有助於提升準確度與影像品質，而女性的掃描量測準確度略高於男性。

除了尺寸的資訊外，本研究設計了四種體型描述工具以量化呈現人體的軀幹型態，包括體積佔有率、輪廓變異、以平面凸多邊形為基礎的輪廓外觀、以橢圓為基礎的輪廓外觀，經評估發現，這些體型描述工具在使用上並不受到掃描姿勢變異的影響。使用因素分析的方法，能夠從這些體型描述工具當中找出具有關鍵描述能力者，並且可以藉由胸、腰、臀等部位的細部特徵說明，有效解釋人體體型的變異。隨後，配合群集分析方法，有助於決定體型的分類群集，得以應用於數位人體模型與服裝設計等領域。

關鍵字

三度空間掃描 ； 人體計測 ； 特徵點辨識 ； 人體尺寸 ； 人體體型

並列摘要

As the demands for customization increase, more detailed understanding and practical use of the size and shape information of human body becomes more and more important. Over the past decade, the 3D scanning technology has opened new opportunities for collecting body sizes and shapes, and related research topics are thus of great concern.

In order to work with 3D scanning data efficiently, this study proposes a method for automated landmarking. After segmentation and initial searches, four algorithms can be employed for locating anatomical landmarks from 3D scanning data, including silhouette analysis, minimum circumference determination, gray-scale analysis, and human-body contour plots. And the method has been proved to be accurate and precise while comparing with traditional palpation works by testing on 189 human subjects.

Based on the results of landmarking, 104 dimensions of the human body can be collected by using approximation methods. A two-stage evaluation was conducted for accessing the performance of scan-derived measurments in terms of accuracy and precision. By scanning 263 human subjects, the precision is high whereas the accuracy is not yet acceptable. Subsequently, by using a mannequin to eliminate the variation caused by human subject, both the accuracy and precision can be improved to meet the expectation. Thus, related counter-measures are critical and beneficial. Further, the effects of arm posture and subject gender were also investigated. It is suggested to adopt the arm posture with palms facing backward to assure higher accuracy and better image quality. Besides, the accuracy tends to be higher for males, and the possible causes should be considered for further improvements.

In addition to size information, the torso shape are also obtained with four shape descriptors, including occupancy, contour variation, convex hull-based contour appearance, and ellipse-based contour appearance. The proposed shape descriptors have been evaluated and were found to be independent of scanning postures. By conducting factor analysis, the key descriptors can be identified to characterize the shape variation in chest, waist, and hip. Further, by performing clustering analysis, the shape clusters can be determined by considering multiple shape descriptors. Numerous applications can be realized, such as digital human modeling and apparel design.

並列關鍵字

3D scanning ； Anthropometry ； landmarking ； body sizes ； body shapes

參考文獻

2. Ben Azouz, Z., Rioux, M, and Lepage, R. (2002) 3D description of the human body shape using Karhunen-Loève Expansion, International Journal of Information Technology, Vol. 8, No. 2, pp. 26-35.

3. Bradtmiller, B. and Gross, M.E. (1999) 3D whole body scans: measurement extraction software validation, Proceedings of International Conference on Digital Human Modeling (paper number: 199-01-1892), Hague.

4. Brooke-Wavell, K., Jones, P.R.M., and West, G.M. (1994) Reliability and repeatability of 3-D body scanner (LASS) measurements compared to anthropometry, Annals of Human Biology, Vol. 21, Issue 6, pp. 571-577.

9. Daanen, H.A.M., Brunsman, M.A., and Robinette, K.M. (1997) Reducing movement artifacts in whole body scanning, Proceedings of International Conference on Recent Advances in 3-D Digital Imaging and Modeling, pp. 262-265, Ottawa.

12. Douros I., Dekker L., and Buxton B. (1999) Reconstruction of the surface of the human body from 3D scanner data using B-splines, SPIE Proceedings, vol. 3640, pp. 234-245, San Jose.