We summarize the different multiview constraints and their characteristics in this section. We are dealing with the class of view-independent constraints that hold good from any view for a configuration of points. Table 1 summarizes the different relations reported in literature.
Constraints on stationary points are essentially the multilinear relations encoded in the Fundamental Matrix, Trilinear Tensor, etc. These are scene-independent. The number of points required to recover these depends on the number of degrees of freedom. Since the points are stationary, time has no relevance.
Levin et al. [11] give view-independent, time-dependent constraints involving 5 points and 8 time instants. In this paper, we gave new view and time independent constraints involving 4 points in 2 views for this case. These work for affine cameras. Both of these are extensible to the perspective camera, but considerably more points or time instants will be required. We also showed how this problem can be reduced to that of stationary lines using the trajectory of the moving points.
This case hasn't been addressed earlier in literature. We derived new view-independent, time-dependent set of constraints for affine cameras that require 4 points and 8 time instants. We also gave time and view independent constraints involving 4 points in 2 views.
Levin et al. derived view-independent constraints for the case of constant angular velocity along an elliptic path.
This problem has not been addressed earlier. We derived constraints by reducing this problem to that of planar contour matching. Our constraints are view and time independent and require a minimum of 1 point for a sufficient number of time instants to evolve a reasonable contour.
