Measuring Bitumen Coverage of Stones

using a Turntable and Specular Reﬂections

Hanna K

all

, Anders Heyden

and Per Lindh

Centre for Mathematical Sciences, Lund University, Lund, Sweden

Peab, Peab Sverige, Helsingborg, Sweden

Keywords:

Segmentation, Classiﬁcation.

Abstract:

The durability of a road is among other factors dependent on the afﬁnity between stones in the top layer and

bitumen that holds the stones together. Poor adherence will cause stones to detach from the surface of the

road more easily. The rolling bottle method is the standard way to determine the afﬁnity between stones and

bitumen. In this test a number of stones covered in bitumen are put in a rolling bottle ﬁlled with water. After

rolling a number of hours the bitumen coverage are estimated by visually investigating the stones. This paper

describes a method for automatic estimation of the degree of bitumen coverage using image analysis instead

of manual inspection. The proposed method is based on the observation that bitumen reﬂects light much better

than raw stones. In this paper we propose a method based on the reﬂections to estimate the degree of bitumen

coverage. The stones are put on a turntable which is illuminated and a camera is placed straight above the

stones. Turning the table will illuminate different sides of the stones and cause reﬂections on different part of

the images. The results are compared to manual inspection and are well in agreement with these.

1 INTRODUCTION

When building roads one wants them to be as lasting

as possible to avoid expensive repairs. Usually the

surface of the road consists of a mixture of stones of

different sizes and a petroleum-based material called

bitumen. To avoid that stones get loose from the pave-

ment the afﬁnity between the stones and bitumen has

to be as good as possible. The afﬁnity is measured by

the rolling bottle method. The goal with this paper is

to improve the manual analysis in this method using

digital image analysis techniques.

1.1 Rolling Bottle Method

The rolling bottle method is a method to investigate

the afﬁnity between stones and bitumen. The stones

are ﬁrst mixed with bitumen so that they are com-

pletely covered in bitumen. After they have been

stored for a few days the stones covered in bitumen

are put in a glass bottle ﬁlled with distilled water.

The glass bottles are then put on a bottle rolling

machine, see Figure 1. On this machine the bottles

are rolling for a couple of hours so that some of the

bitumen gets teared off from the stones. After rolling

a few hours the bottle is removed from the machine to

Figure 1: A bottle rolling machine.

estimate the degree of bitumen coverage. The stones

are put on a piece of silicon coated paper and two ex-

perienced laboratory assistants are visually observing

the stones in order to estimate the degree of bitumen

coverage.

A problem with current state of the art is that it

is not objective, two different labs can get different

result since the degree of bitumen coverage is esti-

mated by different laboratory assistants in different

labs. It is also very hard to make a correct estimation

and the accuracy of the estimations are not sufﬁcient.

The purpose of this project is to improve the estima-

tion by taking photographs of the rolled stones and

then use digital image analysis techniques to analyze

the stones. This would make the method more objec-

tive since the same computer program can be used in

different labs.

333

Källén H., Heyden A. and Lindh P..

Measuring Bitumen Coverage of Stones using a Turntable and Specular Reﬂections.

DOI: 10.5220/0004291103330337

In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2013), pages 333-337

ISBN: 978-989-8565-47-1

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

1.2 Previous Work

In (Merusi et al., 2010), an algorithm for trying to es-

timate the degree of bitumen coverage by using image

analysis has been developed. In the proposed method,

a cyan-colored background for easy segmentation of

the background has been used. To avoid sparkles

and reﬂections in the image a cyan-colored truncated

cone, with the camera in one of the bases, is used. To

classify pixels either as stones or bitumen, a princi-

pal component analysis was implemented. Using the

ﬁrst component the images were thresholded and pix-

els below the threshold were classiﬁed as bitumen.

A more advanced method for estimating the de-

gree of bitumen coverage was suggested by (Well-

ner et al., 2011). To avoid reﬂections in the bitu-

men surface, the stones are put in a crystallization

dish where they were covered with distilled water. A

plastic cylinder were put around the aggregates and

illuminated from outside to ensure diffuse lightening

to prevent shadows to occur. A probability based seg-

mentation method was used for segmenting the im-

ages. To train parameters in the classiﬁer, reference

images on the background, the raw aggregates and ag-

gregates completely covered in bitumen were used.

Both these methods rely on a difference in appear-

ance between the aggregates and bitumen. In this pa-

per we focus on the more difﬁcult problem when the

color of the stones are very similar to the color of bi-

tumen.

Concerning segmentation there is a vast literature

describing several different segmentation methods.

The ﬁrst methods were based on thresholding and re-

gion growing techniques. Also methods from math-

ematical morphology were frequently used (opening,

closing, etc.) in order to smoothen out the contours.

The starting point of modern segmentation methods,

based on variational formulations, was the introduc-

tion of active contours, so called snakes, see (Kass

et al., 1987).

A development of active contours to more general

level-sets was done by Osher and Sethian in (Osher

and Sethian, 1988) and (Osher and Fedkiw, 2003).

The main advantage of the level-set representation

is the ﬂexibility to change topology and improved

numerical methods. A faster version of level-sets,

so called fast marching, was presented in (Sethian,

1996).

Another approach to segmentation based on vari-

ational methods is the so called area based methods.

The pioneering work, the Chan-Vese method, is based

on the Mumford-Shah functional, see (Chan and Vese,

2001). Yet, the main drawback of those methods is

the existence of local minima due to non-convexity of

the energy functionals. Minimizing those functionals

by gradient descent methods makes the initialization

critical. A number of methods have been proposed

to ﬁnd global minima such as (Appleton and Talbot,

2006; Chan et al., 2006).

A new development into discrete methods, based

on graph-theory, is the so called graph-cut meth-

ods, introduced by Boykov, Kolmogorov and others,

(Boykov and Kolmogorov, 2001; Boykov and Kol-

mogorov, 2004; Kolmogorov and Zabih, 2004). The

main advantage of these methods is that they can

guarantee that the solution reaches the global mini-

mum and they are usually very fast.

2 METHODS FOR ESTIMATING

THE DEGREE OF BITUMEN

COVERAGE

A problem when trying to take images of stones cov-

ered in bitumen is that we often get specular reﬂec-

tions in the bitumen. The idea in this paper is to in-

stead of trying to avoid the specular reﬂections we try

to use it for segmenting the images. For that reason

we want to take several images, typically 20-30, with

light from all possible directions. In practice it turns

up to be more practical to place stones on a turntable

which we turned a bit between images than to place a

high number of light sources around the scene.

Our system for analyzing the images then consists

of three parts. First we have to register the images

to each other. After registration we segment the fore-

ground, stones, from the background using all images.

Last, for the pixels classiﬁed as foreground we esti-

mate the degree of bitumen coverage by using a prob-

ability based classiﬁcation method.

2.1 Experimental Setup

The setup used to take images can be seen in Figure 2.

In the setup we have one camera, one light source and

one turntable. The camera is placed straight above the

turntable and facing downwards, looking at the stones

from above. Beside the camera we have a light source

that illuminates the stones from one direction. By

turning the turntable we get light from many more di-

rections. To easier segment the stones from the back-

ground we use a blue background on the turntable.

Figure 3 shows some examples of images that we

get from our setup, these stones are completely cov-

ered in bitumen.

VISAPP2013-InternationalConferenceonComputerVisionTheoryandApplications

334

Figure 2: The experimental setup for taking the pictures.

The camera is looking straight down to the turntable, the

lamp gives light from one direction but turning the turntable

different sides of the stones will be illuminated.

Figure 3: Example of images, the original images before

transformation.

2.2 Registration and Segmentation of

Stones from Background

To be able to use the images we have to register them

to each other. This is done by extracting some corre-

sponding key points in all images and compute a ho-

mography from all images to some reference image.

The homography is a 3 × 3 matrix H so that

λy = Hx, (1)

where H is the homography, λ a scaling factor, x is the

point in the reference image and y is the correspond-

ing point in the image that we want to transform, x

and y are given in homogeneous coordinates.

Then the images are transformed according to the

homography associated with the current image. Fig-

ure 4 shows the same images as Figure 3 after the

transformations.

When the images are transformed we want to ﬁnd

out which part of the image that is stone and which

part is background. Since the shadows are quite sharp

Figure 4: Example of images, the images after transforma-

tion.

Figure 5: The mean image used to segment foreground,

stones, from background.

in the images we take a mean image of all the im-

ages and use that for segmentation. The mean image

can be seen in Figure 5, now the shadows are much

smoother. The segmentation is done by thresholding

in the blue channel of the image. The threshold is

chosen manually, which is not crucial for the segmen-

tation result.

2.3 Estimation of the Degree of Bitumen

Coverage

To estimate the degree of bitumen coverage we look

at the difference between the highest value for a pixel

through all images and the lowest value. If there are

any specular reﬂections in any of the images this dif-

ference will be high. The difference image for stones

completely covered in bitumen can be seen in Fig-

ure 6, as can be seen in the image we do not get re-

ﬂections everywhere. We use some reference images

with stones covered in bitumen and the raw stones to

build histograms for the difference of the highest and

the lowest value. These histograms are then used to

ﬁnd a probability function, that tells how likely a pixel

with a certain difference is to be bitumen and stone

respectively. The histograms are normalized so that

they sum up to 1. The probability that a pixel with

MeasuringBitumenCoverageofStonesusingaTurntableandSpecularReflections

335

Figure 6: The difference image.

Figure 7: Histograms and probability functions for stone

material A and B.

intensity i is bitumen can be calculated by

(i) =

(i)

(i) + h

(i)

, (2)

where P

(i) is the probability that a pixel with inten-

sity i is bitumen, h

(i) is the value of the histogram for

bitumen pixels with intensity i and h

(i) is the is the

value of the histogram for stone pixels with intensity

Figure 7 shows the histograms and the probability

functions for two different stone materials. The blue

curves show the curves for bitumen and the red curves

show the curves for stone.

To estimate the degree of bitumen coverage for

stones that are partly covered in bitumen the differ-

ences for all pixels are computed. The image is also

segmented into foreground and background. For all

the foreground pixels, the probability that a pixel is

bitumen is calculated. Figure 8 shows an image of the

probabilities that pixels is bitumen, white means that

a pixels is very likely to be bitumen and black pixels

are very unlikely to be bitumen. Then the degree of

bitumen coverage is estimated by

dbc =

∑

P(i is bitumen)

, (3)

Figure 8: Probability image for being bitumen, white indi-

cates high probability and black low.

where dbc is the degree of bitumen coverage,

P(i is bitumen) is the probability that pixel number i

is bitumen and N is the total number of pixels. Only

pixels that were classiﬁed as foreground are consid-

ered.

3 EXPERIMENTS AND RESULTS

The method has been tested for two different stone

materials, one dark and one lighter. The results from

the image analysis have been compared with the vi-

sual investigation by experienced laboratory person-

nel. Table 1 shows the result for the two materials.

The results are close to the visual estimations by the

laboratory assistant, but the visual estimation could

also deviate from the true answer.

Table 1: The degree of bitumen coverage for the different

stone materials estimated both by the image analysis system

and by visual inspection.

degree of bitumen coverage

image analysis manual inspection

material A 46.8 % 50 %

material B 29.2 % 35 %

4 CONCLUSIONS AND FUTURE

WORK

With this method we can automatically compute the

degree of bitumen coverage even for stone materials

with a darker color close to the color of bitumen. We

still have to work a bit on the lightening arrangement

to ensure to get more specular reﬂections in the im-

ages.

VISAPP2013-InternationalConferenceonComputerVisionTheoryandApplications

336

ACKNOWLEDGEMENTS

This work was founded by SBUF. We also want to

thank PEAB for supplying images and stone material.

REFERENCES

Appleton, B. and Talbot, H. (2006). Globally minimal sur-

faces by continuous maximal ﬂow. pami, 28(1):106–

118.

Boykov, Y. and Kolmogorov, V. (2001). Fast approximate

energy minimization via graph cuts. Pattern Analy-

sis and Machine Intelligence, IEEE Transactions on,

23(11):1222–1239.

Boykov, Y. and Kolmogorov, V. (2004). An experimental

comparison of min-cut/max- ﬂow algorithms for en-

ergy minimization in vision. Pattern Analysis and Ma-

chine Intelligence, IEEE Transactions on, 26(9):1124

–1137.

Chan, T. and Vese, L. (2001). Active contours with-

out edges. IEEE Transactions on Image Processing,

10(2):266–277.

Chan, T. F., Esedoglu, S., and Nikolova, M. (2006). Al-

gorithms for ﬁnding global minimizers of image seg-

mentation and denoising models. SIAM Journal of Ap-

plied Mathematics, 66(5):1632–1648.

Kass, M., Witkin, A., and Terzopoulos, D. (1987). Snakes:

Active contour models. Int. J. Computer Vision,

1(4):321–331.

Kolmogorov, V. and Zabih, R. (2004). What energy func-

tions can be minimized via graph cuts. Pattern Anal-

ysis and Machine Intelligence, IEEE Transactions on,

26(2):147–159.

Merusi, F., Caruso, A., Roncella, R., and Giuliani, F.

(2010). Moisture susceptibility and stripping resis-

tance of asphalt mixtures modiﬁed with different syn-

thetic waxes. Transportation Research Record: Jour-

nal of the Transportation Research Board, 2180(-

1):110–120.

Osher, S. and Fedkiw, R. (2003). Level Set Methods and Dy-

namic Implicit Surfaces. Springer-Verlag, New York.

Osher, S. and Sethian, J. A. (1988). Fronts propagating

with curvature-dependent speed: Algorithms based on

Hamilton-Jacobi formulations. Journal of Computa-

tional Physics, 79:12–49.

Sethian, J. (1996). A fast marching level set method for

monotonically advancing fronts. Proc. Nat. Acad.

Sci., 93(4):1591–1595.

Wellner, F., Kayser, S., Marschke, L., Schlesinger, D., Mor-

genstern, A., and Schulze, C. (2011). Optimierung

der afﬁnit

atspr

ufung - verbesserung der pr

azision

der pr

ufung zur besimmung des haftverhaltens zwis-

chen groben gesteinsk

ornen und bitumen. Forschung

Straßenbau und Straßenverkehrstechnik.

MeasuringBitumenCoverageofStonesusingaTurntableandSpecularReflections

337