|
Figure 3:
A fast multiresolution algorithm for automatic detection/recognition of traffic signs. |
We have made several significant extensions to our methodology for
constructing multiresolution models for random fields, resulting in
models that yield both visually and quantifiably accurate models and
reconstructions for a wide variety of phenomena. We have
demonstrated the method constructing low-order multiresolution models
from complex random fields. This modeling methodology has also been
applied to multiresolution imagery.
The models have been used as part of a discrimination
test, with results that show considerable improvement over current
baseline system. We
seek a framework in which to answer efficiently two key questions:
(1) When do current object recognition,
estimation and tracking systems break? (2) Where are the key hard parts
of the recognition, estimation and tracking problem?
Fig. 3 shows an automatic recognition
of traffic sign for real-time and autonomous navigation.