Reducing search space traversal in 3D scene reconstruction

Three-dimensional scene recovery from two-dimensional image data is a challenging task. Typical methods compute two-dimensional image feature correspondences between image frames. However, these methods introduce errors as three-dimensional warping of image elements is not considered. Further, no previous method presented exploits the strong correlation between three-dimensional scene elements in adjacent video frames. In this work, we describe a simple three-dimensional scene recovery method which is extended using a novel extension exploiting temporal information, increasing efficiency of the method. The proposed method is applied to the recovery of outdoor scenes. Comparing with other, more traditional, three-dimensional recovery methods, the proposed method provides more accurate results and the addition of temporal information is shown to speed up execution without reducing accuracy.