Page 160 - Introduction to Autonomous Mobile Robots

145
Perception
V
and together capture its chrominance. Thus, a bounding box expressed in YUV space
can achieve greater stability with respect to changes in illumination than is possible in
RGB space.
The CMVision color sensor achieves a resolution of 160 x 120 and returns, for each
object detected, a bounding box and a centroid. The software for CMVision is available
freely with a Gnu Public License at [161].
Key performance bottlenecks for both the CMVision software, the CMUcam hardware
system, and the Cognachrome hardware system continue to be the quality of imaging chips
and available computational speed. As significant advances are made on these frontiers one
can expect packaged vision systems to witness tremendous performance improvements.

4.2 Representing Uncertainty

In section 4.1.2 we presented a terminology for describing the performance characteristics
of a sensor. As mentioned there, sensors are imperfect devices with errors of both system-
atic and random nature. Random errors, in particular, cannot be corrected, and so they rep-
resent atomic levels of sensor uncertainty.
But when you build a mobile robot, you combine information from many sensors, even
using the same sensors repeatedly, over time, to possibly build a model of the environment.
How can we scale up, from characterizing the uncertainty of a single sensor to the uncer-
tainty of the resulting robot system?
We begin by presenting a statistical representation for the random error associated with
an individual sensor [12]. With a quantitative tool in hand, the standard Gaussian uncer-
tainty model can be presented and evaluated. Finally, we present a framework for comput-
ing the uncertainty of conclusions drawn from a set of quantifiably uncertain
measurements, known as the error propagation law.

4.2.1 Statistical representation
We have already defined error as the difference between a sensor measurement and the true
value. From a statistical point of view, we wish to characterize the error of a sensor, not for
one specific measurement but for any measurement. Let us formulate the problem of sens-
n
ing as an estimation problem. The sensor has taken a set of measurements with values
ρ i . The goal is to characterize the estimate of the true value EX[] given these measure-
ments:

,
,
,
(
EX[] = g ρ ρ … ρ ) (4.50)
n
1
2