Herebellow, we describe the contribution of RMA/SIC to Hispars project - Euclid RTP6.2.
In this project an image processing framework and three demonstrators (naval, air, ground)
were developed. RMA/SIC was in charge of a part of the ground range demonstrator, namingly
the identification of armoured vehicle at short range using IR images.
The goal is to identify armoured vehicles from a short range infra red image.
A typical input could be :
for which the system answer should be : amx30
- We are dealing with short range images (about 150x150 pixels) which implies that
a lot of information is present. We could believe that all that information
leads to an easy problem (a man is able to identify the vehicle with a very high level
of confidence) but the problem is to extract the pertinent information.
A direct extraction and understanding of the whole information seems foolish.
- We are dealing with 3D objects which may produce a huge number of different
images. This increases the problem of extracting the pertinent information
because that information depends on the view.
Description
- Location of the vehicle in space (position and orientation). The output is
an axis system fitted to the vehicle.
On the following example, the defined axis system
is drawn in green and the found one in red
(for Y and Z axis, only the direction is searched).
- Hypothesise a vehicle (all vehicles of the database or candidates returned by first module)
- Search at known (axis system, hypothesised vehicle and database) for details.
The following image presents an example of searched details.
- Compute the probability the image
represents the hypothesised vehicle.
- Classify as most probable vehicle or unknown (max probability too small)
Schema
This research was supported by the 'Cabinet des Technologies, de la Recherche et de l'Energie' of
the Walloon Region of Belgium and BATS in the scope of Hispars project - Euclid RTP6.2
