Next: Deliverables
Up: Perception and Action Hierarchies
Previous: Hierarchical Vision
One of the prime uses of vision in the UCAV will be for visual
servoing and tracking for several applications, such as
- Tracking and acquiring targets
- Landing on the ship, possibly in high sea states
In each of these applications, the control loop will need to be closed
around the vision sensor. We have had experience with performing
visual servoing for cars and also for mobile helicopter
platforms. There are two primary approaches to visual servoing:
- Tracking in the World or 3-D coordinates. Here, we use
multiple camera views and motion flow equations to build a model of
the moving object in 3-d space. We design control laws to track this
motion. The problem is subtle, since the control law has to be a look
ahead control law with a typical look ahead time of several
milliseconds to allow for the processing time of the vision
algorithms. This approach will be important to the target tracking and
acquisition mentioned above.
- Tracking in the Image plane (2-D). Here, we track the
object (such as the surface of the landing pad) in the image frame, so
as to enable the camera platform to make contact (land) on a curve or
a surface patch in the image plane. The mathematical formulation of
this involves control of the space of curves in the image plane, and
some new and novel techniques, which we are currently experimenting
with on our laboratory helicopter model. This is relevant to the
landing and take-off scenario described above.
S Sastry
Sun Aug 9 16:58:51 PDT 1998