icccbe 2010 Nottingham University Press Proceedings of the International Conference on Computing in Civil and Building Engineering W Tizani (Editor) Development of system and algorithm for evaluating defect level in architectural work Chollada Laofor & Vachara Peansupap Department of Civil Engineering, Faculty of Engineering, Chulalongkorn University, Thailand Abstract The visual quality inspection in architectural work needs the method to minimize the problem of confliction about acceptable level of defect in construction project. By its nature, the aesthetic issue in architectural work is subjective judgment, which depends on each personal perception. This research paper presents methodology of system development and algorithms for evaluating defect level. The research applies the digital image processing and knowledge management concepts. Research objectives are to support the visual quality inspection and replace the making sense of human vision in subjective judgment. The inspection of tiling work is used to explain the methodology and present example of algorithms for checking the completion of work such as inspection of the distance between neighbouring tiles and the angle of intersecting straight lines. The expected benefits are to know the actual quantity of defect, increase the reliability of visual inspection on construction products and reduce the conflict with several project participants who are involved in evaluating quality via visual inspection. Keywords: visual quality inspection, image processing, defect evaluation, knowledge management 1 Introduction The current practice of quality inspection still encounters problems about the acceptable judgment of defect level. The construction project may not meet the customer requirements although it may be completed on time under budget and functions. Due to different customer perception, the quality in construction work can have several levels. Therefore, inspectors need to know the customer requirement standard for accurate judgment on quality of work. By its nature, quality evaluation in a construction work can be divided into two attributes that consist of measurable and subjective attributes. The measurable attributes are related to material, construction and functional requirements. These conform to contract document that includes the standard, sampling and specification of construction works. As these can be measured, they are frequently used the mechanical instruments to enhance the sensory input for the human judgments. Instruments such as gages are used to determine thread sizes, gap thicknesses, angles between parts, hole depths, and weld features. Thus, the judgment on quality of work can be evaluated by comparing between measurement result and specification. In contrast, the subjective attribute relates to aesthetic issue especially in architectural works. The content in the contract document uncovers about this issue. The satisfaction of aesthetic level depends on customer perceptions which are difficult to identify in contract document. The inspectors use the individual experiences in judgment on quality of work. It has multiple-standards
and unreliability depending on individual experience of each person. These lead to conflict with several project participants who involve in evaluating quality of work such as customer and contractor (Toakley and Marosszeky, 2003). Therefore, the quality evaluation of subjective attribute needs the method to minimize these problems. 2 Digital image processing in construction quality inspection Digital mage processing means the computerized process which is especially used to improve the image quality into suitable form with objective. For example, low and mid levels involve primitive operation such as noise reduction, contrast enhancement, image sharpening, converting into binary image, edge detection or interpretation of image feature. High level involves making sense of an ensemble of recognized object and image analysis by using neural network or fuzzy logic theories (Gonzalez et al. 2004; Yodrayub 2007). Interest in digital image processing methods stems from two principal application areas. These are the improvement of pictorial information for human interpretation and the processing of scene data for autonomous machine perception (Gonzalez and Woods, 1992). The construction industry illustrates many previous research works which are studied to overcome the limitations of subjective evaluation in visual quality inspection by human inspector (Georgopoulos et al. 1995; Lee 2004; Lee et al. 2006). They developed an automatic procedure replacement by using the computer vision and image processing technologies to automate the process. These attempts are to determine the defects of road infrastructure by using digital image processing. The result of the process can help to optimize infrastructure maintenance strategies in operation stage. The digital image processing (DIP) is a popular information technology in infrastructure field as Lee, Chang & Skibniewski (2006) who studied the inspection of the deterioration of steel bridge coating. Georgopoulos, Loizos & Flouda (1995), and Lee (2004) proposed the information technology for inspecting the pavement crack. Therefore, the use of image processing can support human judgments in quantifying and evaluating construction defects. Although previous researches aim to support human subjective judgments by using image processing to detect and quantify construction defects, few researches focus on evaluating the intensive defect level in subjective attribute of aesthetic issue during the construction stage. The acceptable level of defect about aesthetic issue in architectural work is difficult to judgment because it depends on each personal perception. These causes lead to confliction. Therefore, we envision challenge to evaluate the intensive defect levels about aesthetic issues using in each organization. 3 Research objectives The research objective is to study an innovation method for supporting the visual quality inspection in architectural work by developing the system of defect level evaluation. The system helps to reduce the human subjective judgment in aesthetic issue which depends on only individual experience. The system development applies the digital image processing and knowledge management concepts. It is more reliable than only the making sense of human vision. Thus, this research paper aims to present methodology of system development and algorithms for evaluating defect level. 4 System architecture According to the traditional visual quality inspection of architectural work, the judgment about acceptable defect level in each defect group must be inspected to meet the quality requirements. For example, the quality requirements of tiling defect groups must be inspected such as specification, lines, joints, distance of gaps. The specification of tiles size and pattern are clearly specified in
contract document. While aesthetic issue of alignment of joint, straight parallel line, uniform distances and level between neighbouring tiles are the subjective attributes. The evaluation uses only the visual inspection that depends on individual perception. Therefore, this process has multiple quality standards and faces with unreliability. From these problems, it is an essential need to propose the system for evaluating defect level. The system can support the judgment of visual quality inspection. It helps to increase the reliability of visual quality inspection by integrating the digital image processing and knowledge management concepts. The digital image processing is applied to detect the defect value by analyzing the image features of interest object which humans solve unwittingly. The defect value can be interpreted in quantitative form and then compared with standard requirement. In addition, knowledge management concept is applied to establish own standard in visual quality inspection about aesthetic issue. Input Pre-processing Data analysis Output Capturing Algorithm of logical and mathematical model Algorithm of logical and mathematical model User interface User interface Image quality enhancement Defect value calculation Evaluation & translation Results of defect level Image transfer Image acquisition Image scale adjustment Image feature analysis Retrieve to compare Capture Knowledge base Figure. 1 System architecture of defect level evaluation (modified from Laofor and Peansupap, 2009) The system architecture of defect level evaluation in Figure 1 can be classified into eight s which are spread in five main stages: (1) input stage (image acquisition ), (2) pre-processing stage (image quality enhancement and image scale adjustment ), (3) data analysis stage (image feature analysis, defect value calculation ), (4) output stage (evaluation and translation ), and (5) capturing stage (knowledge base ). The data flow starts from the data input stage of image acquisition. It uses the digital camera to capture picture and transfers it into computer for processing. At this stage, the picture is adjusted the image quality and scale into a suitable form by using image quality enhancement and image scale adjustment s. Next, the data analysis stage analyzes the defect feature by capturing the image features such as regions, edges, scale, interest points and texture. After each feature is transformed into measurable dimensions, these dimensions are analyzed the defect value by using the algorithms of image processing which are characteristics of mathematical logic. Finally, the evaluation and translation compares the measurement result from previous with the defect classification in quality knowledge base. Then, output stage of the system can be justified the level of the defect. 5 Methodology The section aims to explain the methodology of system development. The methodology is divided into three main parts: (1) defect value measurement, (2) evaluation and translation and (3) knowledge base of quality standard. The system uses Visual Basic 6 program for developing image processing s. We also use MS Access 2003 for assessing the defect classification and storing quality
standard for knowledge base. The case study of tiling inspection is chosen to explain the methodology and algorithms in section 5 and 6. There are details as follows. 5.1 Defect value measurement The defect value measurement uses the digital image processing for detecting and interpreting the defect value from image feature analysis. The methodology can explain in each that is component of defect value measurement. 5.1.1 Image acquisition The image contains the specific defect group for inspecting in each quality requirement. Under tiling inspection, the tile must be set in straight parallel lines. The distances between neighbouring tiles must be uniform. The neighbouring tiles have to be on the same level. The glue has to be spread uniformly over the entire back of the tile. The tile must be pressed evenly against the floor (Navon, 2000). Our system did not intend to evaluate all quality requirements by using only image processing technique. But we focus on supporting visual quality evaluation in some quality requirements. Therefore, this case study can use the system to inspect the straight parallel lines, uniform of distance between neighbouring tiles. The image acquisition must show the line, joint and level. The images should be taken in the same environment condition such as camera specification, pixels size (640x480), camera distance and others. The representative images of system development should be several intensive defect levels from housing projects. 5.1.2 Image quality enhancement The image enhancement aims to improve the quality of digital image into suitable form for easier analysis by using the digital image processing method. The image acquisition of tiling inspection needs to reduce the noise and convert colour image in Figure 2(a) to binary image (black = 0 and white = 1) in Figure 2(b) by threshold techniques for clarity of gap lines in image analysis. The program must identify the suitable threshold value (0-255). If pixel value is lower than threshold value, this pixel value will convert to 0 (black). In contrast, if pixel value is more than threshold value, this pixel value will convert to 1 (white). (a) 255 255 255 255 255 255 255 255 (b) 1 1 1 1 1 1 1 1 Figure. 2 Image quality enhancement. 5.1.3 Image scale adjustment The image acquisition needs to adjust the image scale unit. Usually, the image acquisition is characteristic of virtual image which is pixel value unit. Thus, this helps to adjust the image unit according to scale ratio of virtual image per real object (pixel per mm.). The scale ratio depends on photography conditions such as camera specification, distance and angle camera and others. For this case study, the distance camera of 0.50 m is suitable distance for clarity of images in digital image processing. The scale ratio of virtual image per real object is 1 pixel equal 1/1.04 mm.
5.1.4 Image feature analysis This used the principle of digital image processing to analyze image features such as regions, edges, scale, interest points and texture. This case study uses the difference of image pixels value between black (0) and white (1) for finding the coordinate of tile edges in both vertical line (V1,V2,..,Vm), horizontal line (H1,H2,..Hn). Moreover, the slope (m) for finding angle of intersecting straight line in each point (P V1H1, P V1H2, P V2H1, P V2H2,.., P VmHn ) can be calculated by using the linear regression equation (See Eq.1). These are shown in Figure 3 after that it can be used to calculate the defect value in next. m = n x j y i - x j y i (linear regression) (1) n( x 2 j ) - ( x j ) 2 Where, m is slope value, x and y are coordinate value on image, n is number of coordinate. V 1 V 2 V 3 V 4 V 5 V 6 V m H 1 H 2 H 3 H 4 H 5 H 6 P V1H1 P V2H1 P V3H1 P V4H1 P V5H1 P V6H1 P V1H2 P V2H2 P V3H2 P P V5H2 P V6H2 V4H2 P V1H3 P V3H3 P V4H3 P V5H3 P V6H3 V2H3 P P V2H4 P V3H4 P V4H4 P V5H4 P V6H4 V1H4 P P V2H5 P V3H5 P V4H5 P V5H5 P V6H5 V1H5 P P V2H6 P V3H6 P V4H6 P V5H6 P V6H6 V1H6 H n Figure. 3 Image feature analysis. 5.1.5 Defect value calculation The defect value shows the intensive defect level from requirement standard by using the algorithm of logical and mathematical model. This case study uses two algorithms in section 6 to check the completion of tiling. First, the algorithm of distance inspection between neighbouring tiles (gap) must be uniform. Second, the algorithm of angle inspection of intersecting straight line must be right angle. 5.2 Evaluation and translation The evaluation and translation is output of system that can be translated into defect level by using the algorithm of logical and mathematical model. The algorithm is used to compare the measurement result from calculation with requirement standard from defect classification in knowledge base. 5.3 Knowledge base of quality standard The knowledge base of quality standard is one part of system to support the evaluation and translation. The main purpose is to store defect value of quality criteria from representative images. To develop the knowledge base of quality standard, representative images should be selected and classified by all project participants who involve in quality inspection process. This process helps the system by retrieving the inspectors perception on quality standard and then can be used to capture the acceptable defect of work from project participants. Next, the representative images are evaluated by the defect value measurement. Thus, the criterion of defect level classification is developed to algorithm logical and mathematical model for evaluating the defect level of new image. This defect classification was stored as the knowledge base of representative quality standard. This acceptable defect in representative image can be used to compare with defect of the following works. The data flow of quality knowledge base is shown in Figure1. This part can be used to create own standard in visual quality inspection about aesthetic issue for improving quality control continuously.
6 Algorithms The system uses the algorithms of logical and mathematical model which are divided into two parts: (1) algorithm of checking completion of tiling and (2) algorithm of evaluation and translation. This paper focuses on the algorithm of checking completion of tiling. The algorithm of checking completion of tiling is divided into two sub-algorithms in Figure 4. First, the algorithm of distance inspection between neighbouring tiles (gap) must be uniform ( X 1 = X 2 = = X n = Y 1 = Y 2 = = Y n ) in Figure 4(a). Second, the algorithm of angle inspection of intersecting straight line must be right angle in Figure 4(b). Figure.4 Algorithms of tiling inspection. 6.1 Algorithm of distance inspection between neighbouring tiles (gap) The defect value is measured from difference of distance between parallel lines and requirement standard (specification) (See Eq.2). n E v = ( X j - s) 2 and E h = ( Y i - s) 2 (2) j=1 n-1 m-1 where, E v is defect value (error) of vertical line, E h is defect value (error) of of horizontal line, X j, is distance between parallel lines of vertical line, Y i is distance between parallel lines of horizontal line, s is standard which is identified in specification, and m, n is sample number of horizontal and vertical lines. 6.2 Algorithm of angle inspection of intersecting straight line This algorithm is used to inspect the right angle of intersecting straight line. The defect value is shown in form of angle value (See Eq.3 to 4) which has an error from right angle (degree = 90 ). Figure. 5 Angle inspection of intersecting straight line. θ = tan -1 m 2 m 1 x 180 or γ = tan -1 m 1 m 2 x 180 (3) 1+ m 1 m 2 pi 1+ m 1 m 2 pi Defect value (θ e ) = 90 - θ = γ - 90 (4) Where, m 1 is slope of horizontal line, m 2 is slope of vertical line, θ is acute angle, γ is obtuse angle, pi = 3.14 and θ e is error value from right angle. m i=1
Figure. 5 Angle inspection of intersecting straight line. 7 Results The results of defect value calculation from representative images are shown in Table 1 and Table 2. The defect value from several representative images can be set as criteria to evaluate the intensive and acceptable defect levels. These are used to be quality standard of organization. Table 1. Results of distance inspection between neighboring tiles (gap). Lines Gaps Defect value Lines Gaps Defect value Vertical E v1 1.914569 Horizontal E h1 1.821254 E v2 1.772944 E h2 1.802932 E v3 1.780168 E h3 1.776646 Table 2. Results of angle inspection of intersecting straight line. Points Defect value Points Defect value Points Defect value Points Defect value P V1H1 0.113259 P V2H1 0.122866 P V3H1 0.357204 P V4H1 0.366875 P V1H2 0.106925 P V2H2 0.116531 P V3H2 0.350870 P V4H2 0.360541 P V1H3 0.003297 P V2H3 0.006308 P V3H3 0.240647 P V4H3 0.250318 P V1H4 0.010851 P V2H4 0.012451 P V3H4 0.233093 P V4H4 0.242764 * These are just samples of all points on image. 8 Conclusions The current construction inspection process always encounters the confliction of acceptable defect level about the aesthetic issue in architectural work. This paper aims to present the methodology of system development and algorithms for evaluating defect level by focus on the part of defect value measurement. The proposed system uses the principle of digital image processing to increase the reliability of visual quality inspection and overcome the limitation of human vision in an analysis of defect features and actual quantity calculation. The case study of tile inspection is chosen to present the methodology and example of algorithms of checking tiling completion. However, the future work needs to study and develop in part of evaluation and translation of defect level. The defect value from representative images are stored to set the evaluated criteria of each defect level. These are used to improve the standard of quality evaluation in each organization to ensure that a quality standard corresponds with the customer requirement. Moreover, the future work needs to study and develop algorithms in other case studies.
References GEORGOPOULOS, A., LOIZOS, A. and FLOUDA, A., 1995. Digital image processing as a tool for pavement distress evaluation. Journal of Photogrammetry and Remote Sensing, 50(1), 23-33. GONZALEZ, R.C. and WOODS, R.E., 1992. Digital image processing. USA: Pearson Prentice-Hall. GONZALEZ, R.C., WOODS, R.E. and et al., 2004. Digital image processing using MATLAB. USA: Pearson Prentice-Hall. LAOFOR, C. and PEANSUPAP, V., 2009. Conceptual framework of defective evaluation system for supporting visual quality inspection in construction work. The 22 nd KKCNN Symposium on Civil Engineering, 2009, Chiangmai, Thailand. LEE, B.J. and LEE, H.D., 2004. Position-invariant neural network for digital pavement crack analysis. Journal of Computer- Aid Civil and Infrastructure Engineering, 19, 105-118. LEE, S., CHANG, L. and SKIBNIEWSKI, M., 2006. Automated recognition of surface defects using digital color image processing. Journal of Automation in Construction, 15(4), 540-549. NAVON, R., 2000. Process and quality control with a video camera for a floor-tilling robot. Journal of Automation in Construction, 10(1), 113-125. YODRAYUB, S., 2007. Digital image processing using visual basic. Bangkok: Technology Promotion Association (Thailand-Japan).