Item

Abstract

Despite many recent advances in optical flow research, many robotic vision
researchers are frustrated by an inability to obtain reliable optical flow
estimates in real-world conditions to apply for real-world tasks. Recently
it has been demonstrated that robust, real-time optical flow is possible
using only standard computing hardware [C93]. One limitation of this
correlation-based algorithm is that it does not give truly real-valued
image velocity measurements. Therefore, it is not obvious that it can be
used for a wide range of robotics vision tasks. One particular application
for optical flow is time-to-contact: based on the equations for the
expansion of the optical flow field it is possible to compute the number of
frames remaining before contact with an observed object. Although the
individual motion measurements of this algorithm are of limited precision,
they can be \em combined in such a manner as to produce remarkably
accurate time-to-contact measurements, which can be produced at real-time
rates, on the order of 6 frames a second on an 50 MHz Sun Sparcstation 20.