Selective Use of Optimal Image Resolution for Depth from Multiple Motions based on Gradient Scheme

摘要

The gradient-based depth from motion method is effective for obtaining a dense depth map. However, the accuracy of the depth map recovered only from two successive images is not so high, and hence, to increase the depth information by tracking corresponding image points through an image sequence is often performed by using, for example, the Kalman filter-like technique. Alternatively, multiple image pairs generated by random small camera rotations around a reference direction can be used for gaining much information of depth without such the tracking procedure. In the framework of this strategy, in this study, to further improve the accuracy, we propose a selective use of the optimal image resolution. The appropriate resolution image is required to have a linear intensity pattern which is the most important supposition for the gradient method often used for dense depth recovery based on the theory of "shape from motion." The performance of our proposal is examined through numerical evaluations using artificial images.