Epipolar Geometry from Temporal Signatures and Dynamic Silhouettes

Epipolar geometry between two images is normally computed by matching feature points between the images. When not enough feature points can be found, e.g. when the camera viewpoints vary greatly, epipolar geometry can be computed from video sequences showing moving objects visible in the two videos. We present a temporal signature for lines: a binary temporal sequence, indicating for each frame the existence of at least one foreground pixel on that line. We show that the temporal signatures of corresponding epipolar lines are very similar. Epipolar geometry can also be found from correspondence of epipolar lines. Correspondence of a pair of epipolar lines can be established using the similarity of their temporal signatures. This search can be limited to a small subset of the pairs of lines, since it can be shown that there exists pairs of epipolar lines each tangent to the silhouette. Compared to existing dynamic silhouette methods, using temporal signatures adds robustness, speed and accuracy. It works even when the silhouette boundaries are noisy, or the silhouette is small due to low resolution or distance of an object from the cameras.