Accurate Estimation of Orientation and Location by Sensor Fusion

在本研究中，提出了一种用于室外环境下摄像机姿态和位置估计的传感器融合方法。在增强现实（AR）系统中，相机（观看者）的姿势和位置是不可或缺的。AR是虚拟现实研究的一个新兴领域，它为用户生成复合视图。它是用户看到的真实场景和计算机生成的虚拟场景的结合，通过附加信息增强场景。在本研究中，提出了一种用于户外AR系统的混合传感系统，可以实现对观看者的精确定位和姿态估计。采用GPS和陀螺传感作为初始估计。利用视觉技术来估计准确的位置和姿态。选取建筑物与天空之间的边界（即天际线）作为图像的衍生特征，进行精细搜索。将图像导出的边界与三维地图合成的边界进行匹配。相似度定义为DP匹配得分。该方法在笔记本电脑上实现，并通过千叶大学校园现场进行评价。场地内建筑的三维模型是使用测量仪器预先构建的。估计的姿态和位置似乎足以说明典型校园场景中的建筑物。在户外，照片经常模糊不清或失焦。尝试了一些图像处理技术来提取天际线，并证明了估计的改进。

In this research, the sensor fusion method for the estimation of camera's pose and location in outdoor environment is developed. The pose and the location of a camera (viewer), are the indispensable in Augmented Reality (AR) systems. AR is a growing area in virtual reality research that generates a composite view for the user. It is a combination of the real scene viewed by the user and a virtual scene generated by the computer that augments the scene with additional information. In this research, the hybrid sensing system for outdoor AR system is proposed, that enables accurate positioning and the pose estimation of the viewer. GPS and gyro sensing are used as the initial estimation. The vision technique is employed for estimate the accurate position and pose. The boundary edges between the buildings and the sky (i.e. skylines) is chosen as the image derived feature, for the fine search. Matching is done between the image derived boundary edges and the synthesized one from 3D maps. The similarity is defined as DP matching score. The method is implemented on laptop PC and evaluation is done through the site, the campus of Chiba University. The three dimensional model of buildings in the site are pre-build using surveying instruments. The estimated pose and position are seem to be sufficient for annotate the buildings in the typical campus scenes. In the outdoor, pictures are often blurred or out of focus. Some image processing techniques are tried to extract the skylines, and the improvements of the estimations has been proved.