A set theoretic approach for time-to-contact estimation in dynamic vision (Q1810574)

This paper addresses the problem of estimating time-to-contact in dynamic vision, one of the main challenges in robot visual navigation and in robot manipulation systems. For mobile robots, the time-to-contact is most important in collision avoidance and in braking. It is well known that differential invariants of the image velocity field (curl, divergence, and shear) can be used to characterize shape changes of objects in a scene because of the relative motion between the observer and the scene. If constant velocity along the optical axis is maintained, the time-to-contact is a function of the area enclosed by the object contour and its time derivative. The authors of this paper propose a new approach based on set membership estimation theory to compute the variables involved in time-to-contact. Errors in the motion model and image measurement noise are incorporated as unknown but bounded disturbances with no assumptions regarding their statistics. The authors' technique permits computation of guaranteed bounds on time-to-contact estimates in finite time, a crucial requirement for practical and robust estimation of time-to-contact. Simulations and experiments illustrate the author's approach. Comparisons to a standard Kalman filter approach are also provided.