Fundamental Matrices from Moving Objects Using Line Motion Barcodes

Abstract

Abstract. Computing the epipolar geometry between cameras with very different viewpoints is often very difficult. The appearance of objects can vary greatly, and it is difficult to find corresponding feature points. Prior methods searched for corresponding epipolar lines using points on the convex hull of the silhouette of a single moving object. These methods fail when the scene includes multiple moving objects. This paper extends previous work to scenes having multiple moving objects by using the "Motion Barcodes", a temporal signature of lines. Corresponding epipolar lines have similar motion barcodes, and candidate pairs of corresponding epipoar lines are found by the similarity of their motion barcodes. As in previous methods we assume that cameras are relatively stationary and that moving objects have already been extracted using background subtraction.

An Epipolar Line from a Single Pixel

“…On average, their number of barcodes is 20x-75x the amount of barcodes we sample. The tables also show the percent of inliers among the line candidates, which are all candidates in our method, and the 1000 line pairs with top matching score in [9]. Our average Symmetric Epipolar Distance (SED) is comparable to [9], and even outperforms their method when applying global refinement, while reducing the time complexity by two orders of magnitude.…”

“…The tables also show the percent of inliers among the line candidates, which are all candidates in our method, and the 1000 line pairs with top matching score in [9]. Our average Symmetric Epipolar Distance (SED) is comparable to [9], and even outperforms their method when applying global refinement, while reducing the time complexity by two orders of magnitude. The run time increase caused by L2 refinement is negligible, and the increase due to L1 refinement depends on the actual number of lines taken into calculation, but it never took more than a second to optimize.…”

“…We tested our method on all camera pairs from both datasets (see Tables 1 and 2 for quantitative comparison with state of the art). Kasten et al [9] sampled a constant number of line barcodes, spaced equally on the image boundaries. On average, their number of barcodes is 20x-75x the amount of barcodes we sample.…”

“…The candidate with the maximal set of inliers is chosen as the epipole. A common definition for inliers in this scenario is the one introduced by [9], which we adopt for our controlled experiments. That being said, when the epipole is inside the image boundaries a simpler approach which works well is measuring whether the perpendicular distance between the epipole and a line is below a certain threshold.…”

Camera Calibration by Global Constraints on the Motion of Silhouettes

“…It is the input to our similarity estimator. It was used in [7,26,8], which also provides a detailed description. The similarity measure is based on a descriptor denoted as motion barcode.…”