头部姿态估计的视觉测量方法研究

As a research focus in the fields of human-computer interaction and biometrics recognition, head pose estimation has been widely applied in many industries. The existing researches on the head pose estimation are mostly carried out when the head is not occluded. But the facial information will not be completely extracted when the head is occluded. It has become an important factor affecting the head pose estimation. The research of this project is carried out for the head pose estimation with occlusion. A new method of head pose estimation with the assistance of circular feature will be proposed based on monocular vision measurement when a person is standing or sitting. The head and the circular feature will have the same motion modes and pose parameters by the research of the particular way of fixing the circular feature on the head. The 3D vision measurement of the circular feature’s pose angles can be carried out by the research of the imaging changes of circular feature and two diameters which are marked on the circular feature. The head motion can be modeled as the rotation around a fixed point of the neck, and then the measurement of head pose angles can be implemented by the computation of rotation angles. The method of head pose estimation can be verified by the experiments with the combination of head model and three freedom degree rotary table. A scientific error model is formed and the general regularity of improving the precision of head pose estimation is revealed. This research will help to improve the practicability and accuracy of head pose estimation, and has important theoretical and practical significance.

头部姿态估计是人机交互与生物特征识别领域的研究热点，在许多行业内有着广泛的应用。现有的头部姿态估计研究多针对头部无遮挡的情况，而头部遮挡将导致面部信息无法完整提取，已经成为影响头部姿态估计的重要因素。本项目将针对遮挡环境下的头部姿态估计问题，基于单目视觉测量原理，研究立姿或坐姿下的一种圆形目标辅助头部姿态估计的新方法。通过研究特定的方式将圆形目标固定于头部，来实现圆形目标与头部具有相同的运动模式和相同的姿态参数；通过研究圆形目标以及标记于其上的两条直径的成像变化情况来实现圆形目标姿态角度的三维视觉测量；将头部运动建模为绕颈部固定点的旋转运动，通过计算旋转角度来实现头部姿态角度的测量；通过头部模型和三自由度转台相结合的实验来进行头部姿态估计方法的验证，建立科学的误差模型，揭示提高头部姿态估计精度的一般性规律。本研究将有助于提高头部姿态估计的实用性和准确性，具有重要的理论和现实意义。

头部姿态估计是人机交互与生物特征识别领域的研究热点，在许多行业内有着广泛的应用。现有的头部姿态估计研究多针对头部无遮挡的情况，而头部遮挡将导致面部信息无法完整提取，已经成为影响头部姿态估计的重要因素。本项目针对遮挡环境下的头部姿态估计问题，基于单目视觉测量原理，研究了立姿或坐姿下的一种圆形目标辅助头部姿态估计的新方法。本项目的主要成果总结如下：1. 研究了将圆形目标固定于头部的方法，以实现圆形目标与头部具有相同的运动模式和相同的姿态参数；2. 研究了圆形目标以及标记于其上的两条直径的成像变化情况，进而实现圆形目标姿态角度的三维视觉测量；3. 研究了人体头部的运动模式，并将头部运动建模为绕颈部固定点的旋转运动，通过计算圆形目标相对于初始位置的旋转角度来实现头部姿态角度的测量；4. 通过头部模型和三自由度转台相结合的实验来进行头部姿态估计方法的验证，建立科学的误差模型，揭示提高头部姿态估计精度的一般性规律；进行了人体头部姿态估计的实验，测量结果显示本项目所提出方法具有较高的精度。本研究有助于提高头部姿态估计的实用性和准确性，具有重要的理论和现实意义。