強化特徵偵測與三維臉部特徵點匹配的頭部姿勢估計法 ; Head Pose Estimaiton based on 2D and 3D Face Feature Points

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/44709

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/44709

Title:	強化特徵偵測與三維臉部特徵點匹配的頭部姿勢估計法;Head Pose Estimaiton based on 2D and 3D Face Feature Points
Authors:	楊小璿;Siao-syuan Yang
Contributors:	資訊工程研究所
Keywords:	頭部姿勢估計;眼睛偵測;嘴巴偵測;三維/二維轉換;3D/2D transformation;mouth detection;eye detection;head pose estimation
Date:	2010-07-27
Issue Date:	2010-12-09 13:53:24 (UTC+8)
Publisher:	國立中央大學
Abstract:	駕駛分心是造成交通事故最重要的因素，因此駕駛昏睡偵測與注意力偵測即為輔助安全駕駛中重要的一環。我們以電腦視覺偵測技術偵測駕駛昏睡程度與頭部方向判斷駕駛的專心程度。根據人臉特徵點在影像與三維空間座標的對應關係，我們提出一個頭部姿勢估測方法來推測出頭部轉動方向。為了估計出準確的頭部姿勢，我們提出強化的特徵偵測方法來提高特徵點位置的準確度。特徵偵測包括眼睛與嘴巴，考慮到行駛中車輛的環境變化，我們的特徵偵測方法可以適應於不同的光影變化的環境與變換的頭部姿勢。在眼睛偵測部分，我們先針對影像上不同照度的區域找出適合的二值化門檻值並且標記連通區域。因為光影或頭部角度的影響，導致眼睛區塊與背景無法順利分割，所以我們接著提出一個可以將突起區塊分割出來的方法，並且以幾何條件的限制篩選出可能是眼睛的物件，最後我們使用支援向量機做眼睛的驗證。嘴巴的位置則是分析眼睛位置的相對關係，再偵測出來的。獲得特徵區塊之後，我們在這些區塊上偵測出特徵點的位置作為估計頭部姿勢的二維資料。我們根據個人化的人臉模型特徵點位置以三維轉換到二維的最小平方誤差估計法求解頭部姿勢參數。The distraction of the driver is the main factor of traffic accident. Hence, the driver drowsiness detection and the driver attention detection are important in safety vehicle system. We use computer vision techniques to detect driver drowsiness or head orientation and determine the concentration of the driver. On the basis of the image coordinates and 3D coordinates of face feature points, we propose an approach of head pose estimation to infer the rotation angles of head. In order to estimate head pose accurately, we enhance feature detection to increase the precision of feature point. Feature detection contains eye and mouth detection. Considering the environment of the moving car, our method can be applied on variant-illumination and different head pose. For eye detection, we find the suitable threshold for binarization at different illumination regions and process connected-component generation. The eye region might be connected with background because of shade or head angle. Hence, we proposed a method that separates raised region from object and apply geometric constraints to obtain eye candidates. Finally, we verify eye using support vector machines. Mouth region can be analyzed according to the location relative to eyes. Then, we detect feature points on the feature region as the 2D data for head pose estimation. We estimate head pose parameters using 3D/2D transformation and least square method according to the feature points of a personal face model.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	706	View/Open

社群 sharing

Loading...