Abstract—This study proposes a novel image-based measurement of crack properties in concrete construction. The

method first retrieves the crack skeleton and profile from the crack map generated by pre-processing the original image.

Then, the crack skeleton is abstracted into a tree structure, and small edges from the crack trunk are removed to calculate

the length of the crack. Finally, Euclidian distance transform is applied on the crack profile to calculate the width of

the crack. The proposed method can identify crack properties automatically and enhance stability, durability, and safety

evaluation of the concrete construction. Validity and accuracy are tested by experiments on crack images of real

oncrete is a complex man-made composite that is widely

used in civil engineering. The behavior of concrete

construction is influenced by environmental affection,

reinforcement, concrete matrix, and the bond among these

features. Surface cracks are the external visible indications of

the behavior of a loaded concrete construction. The appearance

of surface cracks and crack propagation may reflect the current

traditional evaluation process. First, the evaluation is time

consuming and difficult to operate. Completing the crack

measurement processes of an entire building may take weeks

for crack data. Moreover, subjective inspection as well as man-

made factors may lead to erroneous judgments [2]. Different

staff may obtain varied measurement results with manual

methods. To measure crack properties, we need to first detect

the crack automatically. Previous studies have stressed the use

of boundary detection [3]. Yamaguchi et al. [6] presented a

percolation model, which is adaptive and could be used in

Cheng et al. [7] presented a crack detection method based on a

threshold operation that uses the mean and standard deviation

of a gray-level concrete building image. The method can

quickly extract the crack from the background, but the threshold

may not be well controlled. Seung et al. [8] proposed a method

that uses an improved Dijkstra algorithm that can enhance the

velocity of calculating the crack length from concrete tunneling.

However, this method is semi-automated and may need to

manually assign a start seed. Yamaguchi et al. [9] used a

calibration line and the brightness of a crack to identify the

[11] used rotation angel information to derive the crack length

and the width. The method can obtain the width of every point,

but the rotation angel information is difficult to retrieve.

Despite their defects, these methods can promote the

development of image-based crack measurement and verify the

feasibility of image-based methods.

Another constraint is the precision of a manual survey, which

is limited by the measurement instrument. Although this

instrument is inexpensive, an expensive instrument is often

required to obtain more accurate results [12-13]. Mechanical

image and then presents a special crack tree to extract the length

and width from the crack. Section 4 describes an application

case of an outdoor concrete wall using the proposed method.

Section 5 further validates the practicability and accuracy of the

method by comparing with existing studies. Finally, Section 6

presents the concluding remarks and future works.

Photogrammetry has been largely applied in the deformation

measurement of concrete buildings since the beginning of the

digital era. As an indirect and non-destructive method, image-

overcome with a fully automatic image-based measurement.

A novel image-based measurement of crack properties is

proposed in concrete construction. The method first retrieves

the crack skeleton and profile from the crack map generated by

pre-processing the original image. The crack skeleton is then

abstracted into a tree structure, and small edges are removed

from the crack trunk to calculate the length of the crack. Finally,

the Euclidian distance transform (EDT) is applied. The

proposed method can identify crack properties automatically.

Compared with manual measurement approaches, the main

on the surface of concrete construction. To improve these

methods, this study measured the crack length and the width

together. The method is different from the above methods,

which only detect the crack length or the width alone.

Moreover, some of the discussed visual methods only focus on

crack detection and disregard the retrieval of crack properties

that could reflect the health condition of the concrete

construction. The proposed method can capture the property

data of surface cracks to provide more information about the

concrete structure. Abstracting the crack skeleton into a tree

3.1 The crack profile using Canny edge detection

Crack images of concrete surface captured by optical

cameras can be obtained, and the crack profile using Canny

edge detection is retrieved as shown in Fig. 1.

Fig. 1 the crack profile using Canny edge detection.

3.2 Retrieval of crack length

This study shows how to use the crack tree to retrieve the

position of the target point and processes the binary image into

a gray image instead of its Euclidean distance to the nearest

also be obtained from this map.

Fig. 3 Definition of crack width according to engineering

4.1 Crack monitoring experiment setup

The complete process of the crack quantification is

demonstrated on a concrete construction surface using the

proposed framework and checking the real accuracy.

Specifically, we monitored an outdoor concrete architecture

using a designed crack image acquisition system. Fig. 4 shows

the setup of the crack image acquisition system. The system

comprises an optical camera, two tripods, and illuminators. A

Nikon camera and an illuminator were used for capturing crack

Calibration is necessary in advance to ease the transfer of the

pixel length or width to the real scale. The condition could be

complicated without a stable device.

Fig. 4. Experiment setup for an outdoor concrete architecture.

One of original crack images captured from the concrete

(a) the original image of concrete surface crack

The profile and skeleton of the given crack can be obtained

from Fig. 6(a) and (b), respectively. Fig. 6(c) exhibits the crack

trunk generated by the proposed edge removal method. The

distance map of the captured crack image with EDT an be

obtained from Fig. 6(d). The crack profile of length versus

width of the given crack is plotted in Fig. 7. These figures

Fig. 6 Visual retrieval of crack length and width: (a) the crack

profile; (b) the crack skeleton; (c) the crack trunk; (d) the

To make the analysis more accurate, we divide the crack

trunk into five segments, as shown in Fig. 8, and determine the

proposed method of the divided crack.

the performance of our method with the results of manual

measurements that use mechanical instruments, such as a

graduated card, a Vernier caliper, a comparator, and a

micrometer. The approach is also compared with the image-

based measurement method proposed by Seung et al. Their

method uses the Dijkstra algorithm to calculate the crack length

after completing the crack extraction. The width of the points is

calculated by composing each area.

Fig. 9 Crack retrieval on five images.

and in obtaining the crack trunk and crack profile.

Crack properties including length and width can be

calculated using image processing by applying a defined rule.

To evaluate the crack measurement results, the crack properties

Table 2 presents the statistical results of the crack properties

using the two methods.

Fig. 10 Absolute measurement errors of crack properties

compared with manual survey: (a) the crack length; (b) the

crack width.

for most civil applications of crack measuring, such as

geological applications and construction works.

This study can measure crack properties and the proposed

method can be applied in many civil applications. The method

starts with crack skeleton and profile detection with pre-

processing. Then, the crack skeleton of a tree structure is

constructed, and the length is calculated. Afterwards,

identifying the crack width is conducted by applying the EDT

to the crack profile. The validity and the accuracy of the

proposed method are ensured through comparisons with manual

surveys and existing image-based methods.

However, the method requires the calibration and this step