Figure 3 : Photographs of the site of Sesimbra, tiles, fragments and semi ordered piles

The tiles on the Sesimbra wreck are coming from a homogeneous cargo but there is some important variation of their morphology from one instance to another. The second important thing is to note that the cargo organization is strongly perturbed, the tiles organization is only visible in some parts of the wreck (Figure 3, left) and even where it is possible to see it the set of tiles is not enough weel organized to be considered as a unique object. Under these conditions we decided to consider tile as individual objects in all the wreck. In order to start the measurement process we need a theoretical model for tile (and also for bricks) with intrinsic parameters and a local reference system.

 
Figure 4 : Theoretical model from Archaeologists

The precise observation of some tiles coming from the Sesimbra wreck shows that the shape can varish from the representation visible in figure 4 to a half trunk of cone. We decide to use a deformable model basically formed by a half trunk of cone with a scale on the OY axis, (see figure 5 below) to fit the shape represented in figure 3.

   

The theoretical primitive corresponding to the tile is a half trunk of cone 9mm thick. The reference system has got its origin in the center of the curved and the 3 axis are taken in the break plane of the tile and in the length direction of it. The OZ vector is the cone axis as it can be seen in the two figures below:

Figure 5 : Reference system for the tile

Figure 6 : Reference system for the tile

   

The possible zones to measure on the tile are: the 4 arcs of circle intrados and extrados, large radius and small radius, the entire area of intrados and extrados are also available zone for measure. In all these 6 zones the operator can choose the number of points he wants to digitize, this insure a good accuracy for the tile restitution but it is time consuming.

As there are thousands of tile on the studied wreck and as more of them are randomized on the site we decide that a high accuracy is not required for the tile restitution and we developed another way to measure and compute the tile with only 3 measured points on one of the 4 arcs of circle.

   

As shown in figure 4 the theoretical model for these tiles contains default values for the intrinsic parameters.

These values are stored in an XML file as a specification of the Roman Tile class defined in Java language.

The reason is that for each kind of Roman Tile it should be possible to obtain different default values but all of these objects belong to the same conceptual class. The differences are only due to the intrinsic parameters default values.

With each default value we have also a threshold which helps to check inconsistency with the theoretical model (for example to check errors in the measurement process).

   <item>
	<hierarchy 	parent="TuileRonde" 
			name="SesimbraRomanTile"/>
		<description creator="Vanessa Loureiro" 
			dateCreation="28 Nov 2007" 
			dimensionUnit="mm" 
			massUnit="kg" 
			volumeUnit="l" 
			instances=""/>
 		<defaultValues
			Height 			= "78.0" 
			heightThreshold		= "10.0" 
			length			= "468.0" 
 			lengthThreshold		= "20.0" 
			width			= "156.0" 
 			widthThreshold		= "10.0" 
			thickness		= "9.0" 
 			thicknessThreshold	= "3.0" 
			radiusL			= "78.0" 
 			radiusLThreshold	= "10.0" 
			radiusS			= "53.0" 
 			radiusSThreshold	= "10.0" 
			mass			= "0.30" 
			volume			= "1.0"/>   
	<!-- 	These dimention are external: radius are extrados radius
		the intrados radius should be obtained as radius - thickness
	-->
	</item>

   

For the tile, a minimum of 3 points is required on the circle but it is possible to pick up a big amount of points. If enough points have been picked up on the curve, the algorithm will try to identify if they are situated on the interior or on the exterior of it for more accuracy. If there are only 3 points it considers the exterior by default. If enough points have been picked up on the curve, the center can be computed by least square. If not we simply take the center of the most distant points as the center of the circle. Then a scale factor is applied in the height direction (OY vector) to approximate the general shape of the tile.

 

Figure 7 : On the left, the measurement interface on the right the tile and the brick after restitution, in VRML

 
Figure 8 : The Tile and the brick, visible in Figure 3, after computation on three points