基于双目视觉的头部姿态估计

Head pose estimation based on binocular vision

基于双目视觉设计一种由人眼在二维及三维空间相对位置的变化估计头部姿态的方法,并将头部姿态分为15种。用光照补偿和肤色模型划分人脸区域的方法减少传统级联分类器易受光照和角度的影响而产生的人眼误检;对双目相机标定以进行立体校正和去畸变,图像经过对极约束、重投影和立体匹配过程得到视差,根据左右图像中对应的人眼位置的视差由三角测量原理算出其深度信息。通过实验证明,本文方法的人眼定位效果和测距精度良好,头部姿态估计准确率高。

Based on binocular vision,a method for estimating the head posture by the change of the relative position of the human eye in two-dimensional and three-dimensional space is designed,and the head posture is divided into 15 kinds.The method of dividing the face area by illumination compensation and skin color model reduces the human eye misdetection caused by the influence of illumination and angle of the traditional cascade classifier;the binocular camera is calibrated for stereo correction and dedistortion,and the image passes through the opposite pole The constrained,re-projection and stereo matching processes obtain parallax,and the depth information is calculated by the triangulation principle according to the parallax of the corresponding human eye position in the left and right images.Through experiments,the human eye positioning effect and ranging precision of the method are good,and the head posture estimation accuracy is high.