On-line Reconstruction of 3-D Structure from Image Sequences Based on Spatio-temporal Information Propagation

1.本项目的成果(1)信息传输向分辨率方向的发展我们提出了一种连续两帧的光流计算算法，该算法可以解决孔径问题和混叠问题。虽然基本思想是应用具有分辨率方向的卡尔曼滤波，但我们对其进行了扩展，以确定在应用卡尔曼滤波之前通常需要知道的参数。通过数值计算，验证了该算法的正确性。(2)信息传播向时间方向的发展利用光流估计和该算法计算出的可靠性，提出了一种基于信息向时间方向传播的深度计算算法。该深度计算算法也是上述具有参数确定功能的卡尔曼滤波算法的具体实现，并且我们在该算法中加入了一种新的深度内插方法。2.未来计划在本项目中，我们研究了强度不变约束完全成立的情况。但是，对于摄影机固定且目标对象移动的情况，此类约束不成立。因此，我们计划扩展本项目中开发的算法，以处理强度变化的情况。

1.Results of this project(1)Development of an information propagation toward a resolution directionWe developed an optical flow computation algorithm using successive two frames, which can solve the aperture problem and the alias problem. Although the basic idea is applying the Kalman filter with a resolution direction, we extended it so as to determine the parameters usually needed to be known before applying the Kalman filter. Through numerical evaluations, it was confirmed that this algorithm correctly works.(2)Development of an information propagation toward a time directionWe developed a depth computation algorithm based on information propagation toward a time direction using the optical flow estimate and its reliability computed by the above algorithm. This depth computation algorithm also is a concrete of the above Kalman filter having parameter determination function, and we incorporated an interpolation method of the depth, which was newly developed, into this algorithm.2.Future planIn this project, we examined the case in which the intensity invariant constraint holds completely. However, for the case that a camera is fixed and target objects moves, such the constraint does not hold. Hence, we are planning to extend the algorithm developed in this project in order to treat the intensity variant situations.