Summary: | The performance of two methods of reconstructing the optic flow from edge motion has been investigated in computer experiments. One is an extremum principle that constrains the optic flow in the image to vary smoothly along a contour feature, and the other is an algorithm that reconstructs the optic flow and the three-dimensional scene and motion parameters on the assumption that the edge features in view lie on a planar surface facet. The noise sensitivity of both methods is similar, and no significant amplification of the noise occurs for noise levels up to approximately 10%. The derivation of the scene and motion parameters, performed only by the second algorithm, is rather more sensitive to noise, however. The effects of perspective on the smoothness constraint are briefly examined. The reconstruction of the optic flow from a rotating feature by the smoothness constraint is affected by vertices in the edge-feature contour, suggesting that the constraint might be improved by including a factor related to the contour curvature.
|