| 摘要 |
Error propagation of errors in the determination of 3D position information from images captured by a moving camera is carefully controlled. A first, second and third image are captured from successive positions. The first and second images are used to determine information representing distances to objects that are visible at image locations in the first and second image. The second and third images are used to determine at least a direction of the translation between capture of the second and third image, without using the first image in the computation of the direction. Next, the direction of translation from the second and third image is combined with the information representing the distances to estimate the amount of translation. In this way, it is prevented that errors from the combination of the first and second image propagate into errors in the determination of the direction of translation between capture of the second and third image. A cross-influence is allowed only when the amount of translation is subsequently determined, but it has been found that this can be done in a way where ill conditioned error propagation can be avoided. |
