利用二維影像人體圍度量測之擬真三維人體模型

Abstract

近幾年來三維影像科技的應用更加的廣泛，尤其是在三維電影與遊戲的應用，在未來的生活中三維影像科技的應用將會無所不在，個人化的三維人體模型將是未來三維影像科技應用的發展重點。因為許多三維影像應用發展將會使用到個人化的人體模型，例如虛擬試衣間、虛擬會議、外科手術評估...等，雖然目前已經有了幾種利用電腦繪圖技術製作個人化人體模型的方法，不過它們所建立的人體模型還是無法充分的描述個人化的差異特徵。所以本論文提出了一個利用真實人體圍度數據來產生擬真三維人體模型的方法，此方法包含了兩個主要的步驟：人體圍度量測與三維人體模型形變。在人體圍度量測方面，本論文提出以二維影像為基礎的人體圍度量測方法，並利用中醫穴道定位法幫助人體圍度量測點定位；在三維人體模型形變方面，本論文提出了廣泛柱狀自由形變法來調變真實三維人體模型，進而產生接近身材圍度量測數據的個人化三維人體模型。為了驗證本論文所提出方法的效能，在本文中以比較人工量測人體圍度來驗證本論文所提出的二維量測方法效能，從結果中發現，此方法有高於96%的準確率。在本文中展示出數個依據二維影像量測身材圍度數據所產生的個人化三維人體模型，並證實了由此方法所產生的三維人體模型與測試者相當接近。

The applications of 3D technology are quite popular in recent years, especially in 3D .lm and games. It is not longer fantasy to imagine that the 3D technology application will be more extensive in future life. The individual 3D human body model is desirable in many 3D applications, such as virtual .tting room, virtual interactive meeting, surgical assessment, etc. Although there are many approaches to create a human body model using the computer gen-erated methods, these models are not adequately represented the real once with complex mor-phology. This dissertation presents an approach to generate a realistic 3D human body model through the use of human anthropometric measurement data. In anthropometric measure-ment, the 2D image based anthropometric measurement is proposed and the location rules of traditional Chinese medicine acupuncture point are used to locate the position of measurement point. In 3D human body model deformation, the extended cylindrical free form deformation is presented to deform the scanned 3D human body model, and then the realistic 3D human body model based on individual body size is generated. For demonstration the performance of 2D image based anthropometric measurement, the results of this system are compared with manual measurement performed by skilled measures, and the results of experiments show that this method has very high accuracy and quite comparable to manual measurement. The gener-ated 3D human body models through the use of 2D image based anthropometric measurement data are demonstrated, and proo.ng that the generated human models are very close to the test subjects.