*7.6.2. Polarimetric investigation of bulk goods with rough surfaces*

One possible discrimination criterion between smooth and rough surfaces is to exploit specific depolarization effects. Rough in this context means that the standard deviation of the outer surface height distribution is in the range of some wavelengths of the operating carrier frequency, i.e. 9 GHz in this case. To obtain results which are independent both of material and shape, extensive measurements were performed.

Four metallic objects with smooth surfaces were used for the investigations, 2 objects with a square cross section of different size, an object with isosceles triangular cross section, and an object with a rectangular cross section. Details are shown in Fig. 44. For comparison with rough surfaces 4 polystyrene bins were built which were filled with bulk goods made of chunky gas concrete, chunky sand-lime brick, medium density fiberboard (MDF) blocks with 0.01 m edge length and M16 × 25 mm screws. Due to the material composition, the polystyrene bin itself has a vanishing radar cross section so that reflections caused by the bin are negligible.

<sup>ï</sup>

middle and on the right side.

recognize contour parts [35].

whether the curve is concave or convex.

the smoothed red curve is plotted.

 !+

<sup>ï</sup>

**Figure 43.** Results of the depolarization investigation of the object with square cross section, with 0.3 m edge length and the bin with same inner dimension, respectively. The depolarization coefficient *P*VH/*P*VV of the smooth surface is depicted on the left. The same coefficient is shown of the bulk goods in the

As mentioned above, the OR algorithms in [50] and [48] yield very robust results with the method of moment invariants and Fourier descriptors, which was proven for images obtained on complete circular tracks around the objects. However, in many cases such complete tracks are not possible as they lead to unfinished object images. To perform an OR also in such situations the method of Curvature Scale Space (CSS) was applied due its ability to robustly

The CSS representation is invariant against rotation of objects as this causes a circular shift of the CSS which has no effect on the recognition process, since the CSS of an object under test is compared by a correlation to the CSS of all reference objects. Moreover, CSS is highly robust against noise as most of its influence is compensated for to some degree by a smoothing Gaussian filter. Another property of the CSS is that it retains the local properties of shapes. Each peak of the CSS corresponds to a concavity or a convexity. A local deformation of the shape causes a change just in the corresponding local contour of the CSS image. Thus, a restricted curve can exactly be found in the CSS of the whole curve. Moreover, the absolute value of a CSS peak indicates the curvature radius, and the algebraic sign of the peak indicates

In Fig. 45 a plane curve with 8 convex or concave parts is drawn in blue. The Gaussian smoothed curve is drawn in red. On the right side of Fig. 45 , the corresponding curvature of

 !+

<sup>ï</sup>

 !+

 %"!\$&

ï

ï

#"\$\*&"!-

 

) %\$(% !ï \$

Cooperative Localization and Object Recognition in Autonomous UWB Sensor Networks 221

ï

ï

ï

ï

#"\$\*&"!-

*7.7.1. Object recognition algorithm with the curvature scale space*

 

ï

ï

""&%'\$

ï

ï

#"\$\*&"!-

 

ï

ï

For depolarization investigations, these 8 objects under test were scanned in both co- as well as in both cross-polarized configurations on a circular track with a radius of 1 m at a 1◦ grid. The depolarization is expressed by the relation of signal powers *P*VH/*P*VV.

In Fig. 43 the results for an object with a square cross section are compared with those of bulk good objects. Expectedly, objects with a smooth surface depolarize least, i.e. about -20 dB in the mean of all measurements. This complies with the 20 dB cross-polarization suppression of the antenna characteristic. The depolarization is least when an edge with predominantly vertical orientation is illuminated. Gas concrete and MDF are the bulk goods which depolarize most. The higher the permittivity is the more the bulky material depolarizes, i.e. sand-lime brick with about 8 dB more and the screws which depolarize most with over 10 dB more than the objects with a flat surface.

So UWB radar seems to be capable of discriminating objects of different materials by the roughness of their surface, subject to the condition that the height deviation is not much smaller than the wavelength. These results highlight the superior capabilities of fully polarimetric systems and recommend their use in future radar systems.
