Image Processing Reference
- it may involve the relative evolution of two objects. In robotic manipulation, for
example, the objective is to model the effector's movement with respect to the object
it has to grip in the scene, in order to be able to determine at any given time the control
laws to apply;
- it may involve the relative movement between the observed object and the
observing sensor. This is the case, for example, with driver assistance systems when
the vehicle is equipped with a sensor capable of detecting and characterizing obstacles;
- finally, it may also be used when the acquisition conditions vary over time and
when it is necessary to modify certain acquisition and processing parameters. For
example, when the lighting conditions of an observed scene vary, it can be useful to
modify a camera's aperture size.
11.2. Temporal constraints
Therefore, the objective of temporal fusion is to determine or evaluate information
with a time-limited validity. The concept of an instant is completely related to the
application. For example, the global warming timescale cannot be compared to the
scale used for locating a plane. In any case, it is necessary to specify the expected
performances depending on how the data is going to be used. During the observation of
dynamic systems, the observation period (sampling period) has to be compatible with
the Shannon theorem: the sampling period must be no greater that half the response
time of the observed system.
This constraint requires the development of specific processing architectures if the
observed system evolves quickly or requires rapid responses, compared to the time it
takes to observe the data [ROM 95].
The processing time leads to a delay that corresponds to the time between the
acquisition of measurements and the transmission of data obtained from processing
these measurements. For example, if the sensor is a CCD camera, the measurement
is an image comprised of pixels which are associated with gray levels. Processing
the image can lead to obtaining data of a higher level such as areas associated with
physical features. The data, obtained after the processing, involves the moment of
observation, i.e. it corresponds to the date of the measurements it is associated with.
Processing causes delays that may or may not be neglected depending on the appli-
cation. For example, if the information is used to determine a control, the delay can
cause an instability, which is a result well known to control engineers; if they are used
to trigger an alarm, the delay can be fatal. A compromise will often have to be made
between the quality of the data and how fast it has to be available.
For each sensor, we need to define the temporal quantities with which it is associ-
ated [ALL 01]. Figure 11.1 represents these quantities: