以深度圖影像與骨架特徵點進行人體重量估測

人體體重是健康的基本指標之一，因為過胖或過瘦的人可能會比正常體重的人還容易產生健康的問題。一個能夠記錄、監測和分析個人體重的方便作法有助於個人健康的照護。此外，一個國家的健康機構對監測人體體重的分佈情況通常會感興趣。人體體重測量的傳統方式都是使用機械式的體重計，但使用傳統體重計去做一項大規模的人體體重分佈研究是很不方便的。在雲端計算環境的考慮下，本論文探討了一個新的方法來進行人體體重估測。利用一個深度體感攝影機自動擷取人體物件的正面及側面輪廓的深度影像，並利用骨架偵測技術找出頭心、雙肩及雙腳踝的像素座標，然後由人體物件將人體骨架面積、骨架長方體和Voxel值等特徵計算出來。最後，以迴歸分析的方式找出這些特徵與人體體重之間的關係，再由假設檢定的過程來驗證迴歸分析的結果。結果指出就統計學而言，所有的特徵都與體種有關。此外，人體Voxel值特徵與人體體重的相關度最高，且經估測體重與實際體重之間的平均誤差率約為2.1%。雖然此數據顯示仍然有改進的空間，不過與傳統機械式體重計相比，本文提出了一個創新的人體體重估測系統並在網際網路的世界中可能有新的潛在應用。

關鍵字

立體像素；人體體重估測；骨架特徵點；深度圖影像；深度體感攝影機

並列摘要

Human weight is a basic index of health because people who are overweight or skinny may suffer from health problems easier than those with a normal weight. A convenient way of recording, monitoring, and analyzing the weight of an individual is useful for personal health care. In addition, a health organization of a country is usually interested in monitoring the weight distribution of the people in the country. Traditionally, a weighting machine such as bathroom scale is used to measure the body weight. A large-scale study on weight distribution of people with traditional weighting machines is inconvenient. With cloud computing environment in mind, this thesis investigates a novel approach to measure body weight. The idea is to use a depth somatosensory camera to capture the depth images of front and side silhouettes of human objects automatically and find the pixel coordinates of head center, shoulders and ankles by using a skeleton detection technology. Then, the human object is characterized by a feature like skeleton area, skeleton cuboid, and human voxel. Finally, a regression analysis approach is used to find the relation between those features and the corresponding human weight. A hypothesis testing process is used to validate the result of regression analysis. It turns out that all features considered are dependent on human weight in statistical sense. Furthermore, the human voxel feature has the highest correlation to human weight and the experiment results show that the error rate between estimated weight and actual weight is 2.1% on average. This performance may still have some room of improving. However, as compared to the traditional weighting machine, this thesis has proposed an innovative human weight measurement system that may have new applications in the world of the Internet.

並列關鍵字

skeleton characteristic ； depth somatosensory camera ； depth image ； voxel ； human weight estimation.

參考文獻

[6] 張錦基，迴歸分析、類神經網路與遺傳演算神經網路於人體脂肪測定及推算之探討，淡江大學資訊工程學系在職專班碩士論文，2009。

[2] A. Myers and J. C. Rosen, “Relation to mental health symptoms, body image, and self-esteem,” International Journal of Obesity, Vol. 23, No. 3, pp. 221-230, 1999.

[3] M.-W. Tsou and J.-T. Liu, “An empirical study of Taiwanese women,” Education Economics Obesity Schooling and Health Knowledge, Vol. 14, No. 1, pp. 89-106, 2006.

[4] 高敏真，台灣肥胖及非肥胖老人的自覺健康狀況及其相關因子之探討，Taiwan Geriatrics & Gerontology，Vol. 3 No. 4，pp. 60-72，2008。

[7] World Health Organization, Obesity and Overweight, 2003.

國際替代計量

以深度圖影像與骨架特徵點進行人體重量估測

全文下載

主題瀏覽