Modified Crack Detection of Sewer Conduit with Low Resolution Images

: Imaging devices of less than 300,000 pixels are mostly used for sewage conduit exploration due to the petty nature of the survey industry in Korea. Particularly, devices of less than 100,000 pixels are still widely used, and the environment for image processing is very bitter. Since the sewage conduit images covered in this study have a very low resolution (240 × 320 = 76,800 pixels), it is very difficult to detect cracks. Because most of the resolution of the sewer conduit images are very low in Korea, this problem of low resolution was selected as the subject of study. Cracks were detected through a total of six steps of improving the crack in Step 2, finding the optimal threshold value in Step 3, and applying an algorithm to detect cracks in Step 5. Cracks were effectively detected by the optimal parameters in Steps 2 and 3 and the user algorithm in Step 5. Despite the very low resolution, the cracked images showed 96.4% accuracy of detection, and the non-cracked images showed 94.5% accuracy. Moreover, the analysis was excellent in quality, also. It is believed that the findings of this study can be effectively used for crack detection with low-resolution images.

This step converts analog images to binary images. For crack detection, it is necessary to find an optimal threshold value. The method proposed by Bradley (2007) was used to compute it, and the parameters include sensitivity and statistical analysis method. Step 3 Make the Binary Image (Find the optimum threshold value) (Parameter : Sensitivity & Statistic) Step 4

Fill Interior Gaps & Dilate the Crack
Step 5 Detect the Crack (Complement the Image) (User Algorithm)

Fig. 3 Flowchart of crack detection
Sensitivity assigns a real number between 0 and 1. If the value is large, there is a risk of including some background pixels. Despite the risk, the larger pixels are regarded as a foreground for the binary conversion. Statistical methods can be selected from mean, median, or Gaussian. Mean is the local average value of the neighborhood, median is the local median of the neighborhood, and Gaussian uses the Gaussian weighted average of the neighborhood as a statistic.
In the example of Table 2, the sensitivity was 1, and the chosen statistic was Gaussian. The parameter study of sensitivity and statistical method is conducted in Chapter 3.

2.4
Step 4: Fill Interior Gaps and Dilate the Crack This step involves filling the inside of the crack and expanding the crack. The method proposed by Soille (1999) was used to fill the inside of the crack, and cracks remain mostly. Any deformation, that is not a crack, can be removed through the following step. The image dilation method proposed by Gonzalez et al. (2009) was applied to highlight the cracks.

2.5
Step 5: Detect the Crack This step determines the presence or absence of cracks, i.e., extracting only the cracks. It determines whether the object filled in the interior in Step 4 is a crack. In this step, a User Algorithm is required and will be explained in detail in Chapter 3. Additionally, it is a step of inverting the black and white of the image to improve readability.

2.6
Step 6: Overlay the Image Finally, it is the step of overlaying the detected crack boundary on the original image. Gonzalez et al. (2009) proposed method was used.

Crack detection parameters and user algorithm
Finally, this step overlays the detected crack boundary on the original image by the method of Gonzalez et al. (2009). Table 3 shows the analysis parameters described in Chapter 2, and Tables 4 to 6 show the analysis results. (a) The result of a crack image refers to the accuracy of finding cracks, and (b) the result of a non-crack image refers to the accuracy of finding non-crack. The smaller the sensitivity, the accuracy of the non-cracked image gets close to 100%, but the accuracy of the cracked image gets closer to 0%. The greater the sensitivity (closer to 1), the higher the accuracy of both cracked and non-cracked images. The reason for this is that, although there is a risk of including some background pixels as the sensitivity increases, more pixels are regarded as a foreground to be binarized. Additionally, it is analyzed that the smaller the value of ClipLimit is, the more pronounced the contrast becomes, and thus showing better results.

User algorithm
User algorithm is used to find cracks in Step 5 of Chapter 2, and it is a step to find only cracks in Step 4 of Table 2. As shown in Table 8, due to the low-resolution image characteristics of the sewer, it is detected like a crack in black on the left and right (the part marked with 'P'). However, this is due to the nature of the image, and it is not a crack. It should be removed. The part marked with 'Q' in Table 8 is barely detected as a crack due to calculation error, but it is not a crack. To remove this error, the crack was detected by judging that the length and the width of the crack should be 9 mm (24 pixels) or more and 3 mm (8 pixels) or more, respectively. Additionally, if the width of the crack was more than half (160 pixels) of 320 pixels, which is the size of the image (240 × 320), it is determined that it is not a crack. Table 9 shows the algorithm for this step.

Results and analysis
Through the crack detection procedure presented in Chapter 2 and the optimum parameters and user algorithm presented in Chapter 3, the crack detection results of the low-resolution sewer image were analyzed.
As a result of crack detection from the sewer images, 53 out of 55 (96.4%) were accurately detected as cracks, and 52 out of 55 (94.5%) were correctly detected as non-cracks. Tables 10 and 11 summarize the results of crack detection from crack and non-crack images, respectively. In Table 10, Bad results refer to detecting non-cracks as cracks, and Good results mean detecting only part of the crack. Excellent results refer to the case of perfect crack detection. Table 11 presents the results of crack detection from non-crack images only. If a crack is detected, when it is a non-crack image, it is considered a bad result. If it is correctly not detected as a crack, then it is an excellent result. Good results are meaningless in this case of non-crack images. Excellent results Images other than the above