基於影像之人體圍度估測系統

Abstract

“人體圍度”意指人們購買衣物所需資訊，如胸圍、腰圍、腿長等。人體圍度量測器(alvascan powered by intellifit)在購物商場正慢慢的被推廣，但現今人體圍度量測的工作必須限定於固定場所，且成本高昂。因此，本論文將提出一個以影像為基礎的人體圍度量測系統，目的在於突破人體圍度量測機的場地限制與減低設備成本。 在本篇論文中，首先利用膚色的特徵與型態學(morphology)的影像處理技巧找出膚色區域，再配合邊緣偵測(edge detection)與連通性(connectivity)的分析，以區域成長法(region growing)正確的分割出人體區域，並對人體未穿著衣物部位做標記。接著，利用模糊推論的方式估測出人體的衣著厚度程度，配合類神經網路訓練出的人體體型，找出在衣著下的身體位置，最後以貝氏曲線來計算出身體圍度，未能計算出的部份也同樣利用類神經網路進行推論。 本論文的主要貢獻主要有二，第一，針對被衣物遮蔽的身體區域，提出一套模糊推論系統(fuzzy inference system, FIS)為核心的衣著厚度推論系統，可以估測出人體在不同衣著下的身體位置，並量測之。第二，只要給於特定部位的正面與側面寬度，便可計算出人體圍度，模擬出不同部位的人體曲線。由實驗數據可得知，拍攝使用者三種固定姿勢的影像，就可以找出人體訂製衣物時所需的圍度資料。

The information of the waistline, chestline, shoulder width, and the length of legs are the essential information when buying clothes. The alvanscan system powered by intellifit is popularized gradually. However, this system works only in some certain place and its price is very expensive. Hence, a new image-based human body shape estimation system is described in this thesis. It can greatly reduce the cost of time and money. In this paper, the skin color region is first segmented by using skin color characteristic and morphology of image processing. Along with the edge detection and connectivity analysis, the correct human body region can be segmented by region growing and the naked parts of human body are marked. Next, the thickness of clothing is estimated by adopting fuzzy inference theorem. Using the body shape trained from the neural network, the real body shape under the clothing can be estimated and determined. The various width and length of human body can be calculated by Bezier curve finally. There are two main contributions in this thesis. First, the body shape under clothing can be estimated by using the proposed clothing thickness estimation system based on fuzzy inference system. Second, the various width and length of human body can be calculated while the information of particular frontal and lateral width is given. Then, the curves of a human can be simulated. From the experiment results, the human body width and length data for buying clothes can be obtained by using three images with fixed poses.