Study of Super Accurate Methods for Geometric Inference from Images

本研究关注的是如何从相机拍摄的图像中估计物体的三维形状等几何信息。我们一直在寻求理论上最准确的方法，并发展它们的应用。我们的研究成果包括一种新的矩阵分解数值方法、一种投影-摄像机系统的标定方法、一种超精确的平面跟踪方法、一种车载摄像机拍摄的城市时空建模和可视化方法、一种车载立体摄像机的标定方法、一种基于测光方法的超精确形状测量方法、一种新的马尔可夫随机场优化方法等。

This study is concerned with the problems of estimating geometric information such as the three dimensional shape of an object from its image(s) captured by a camera. We have sought for the theoretically most accurate methods along with development of their applications. Our achievements include a new numerical method for matrix factorization, a calibration method for projector-camera systems, a super-accurate planar tracking method, a method for spatiotemporal modeling and visualization of a city from its images captured by a vehicle-mounted camera, a calibration method for vehicle-mounted stereo cameras, a super-accurate shape measuring method based on the integration with photometric methods, new optimization methods for Markov random fields, etc.