Determination of displacements and deformations of the structural systems by means of template matching method

In this master thesis, a cost-effective and wireless image processing technique has been developed as an alternative to the conventional measurement methods using wired cables in structural systems or wireless (remote) measurement methods having high costs. Within the scope of the study, theoretical and experimental studies have been carried out. An algorithm has been established in the light of the literature review and several experimental works carried out in Istanbul Technical University (İTÜ) Structural and Earthquake Engineering Laboratory. Besides the classical measurement methods, the developed algorithm was used and the obtained results were compared. In the first part of the study, a short information about digital image processing, the purpose and the scope of the study and a brief summary about template matching method (used) are given. Usage areas of digital image processing such as medical applications, face detection and object tracking etc..., the reasons of preference and usability on structural systems are mentioned. Additionaly, comparisions between developed method and conventional methods are given to understand why image processing techniques should be used in different study areas. Digital image processing is a technique of processing digital images which are transferred to a computer by means of the written algorithms. Digital image processing, one of the branches of the signal processing, allows many applications to be applied together. The displacement, velocity, acceleration or deformation values of structural elements are measured by conventional methods to make structural health monitoring on real structures and to observe structural models. There is a need for new wireless measurement system due to classical methods which have high cost, have a requirement of a lot of cabling works and have effects on properties of specimen using in laboratory. There are studies showing that digital image processing, which is preferred for use in different work areas, is also applicable on structural systems.In this study, it is tried to make alternative method to conventional methods by using template matching algorithm on Python software. In the second chapter, after the definitions of digital image, pixel and digital image processing are given, it is tried to categorize the digital image processing methods which can be used on structural systems. All categorized methods are summarized. The literature review of digital image processing methods is also included in this section. When three dimensional objects are transmitted to two dimensional area by using camera recording, each digital image is transferred as a matrix of elements in row and column indices. Each element in this matrix contains color, texture, brightness etc. informations about the point in which it is located. Each element in the matrix is called as pixel. In this study, since the displacement and strain magnitudes are wanted to be found by using a digital image proccessing technique, pixel has an important role for the process. The magnitudes can be found with the pixel attitudes by utilizing object detection techniques or object tracking techniques. Object detection techniques have more complicated algorithm than object tracking techniques because they perform object tracking after object detection. There are several object detection methods. In this study, these techniques are tried to be collected under 3 main heading; video segmentation, background elemination and optical flow. Each of the methods has its own subcategories and difficulties. There is a brief summary of each method in this thesis. A considerable number of approaches are proposed in the literature about object tracking. The difference between object tracking methods and object detection methods is that the information of object to be tracked is given by the user. In this study, these approaches are grouped under three different headings; point-based tracking, kernel-based tracking and silhouette-based tracking. A brief summary of each approach is given. In the third chapter of thesis, detailed information about the algorithm developed by the proposed template matching method is given. Proposed template matching methods in opencv library, the selected method among these methods and their mathematical operations are mentioned. Information is given on filters used to remove erroneous data due to noise and imaginary impairments captured in video images. Python, an open-source programming language, has been used to obtain motion in the x and y directions of some specific points on structural models which are under the static/dynamic loads. The libraries like openCV, numpy, scipy, matplotlib and math are used on Python to perform mathematical operations, to open and to process video record, to use filters, to transfer data to different programs. The template matching method is used to find the system properties of the structures on the images.To determine the correlation map in the proccess of template matching, openCV library offers 6 different calculation methods: TM_CCOEFF, TM_CCOEFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_SQDIFF and TM_SQDIFF_NORMED. The method used in algorithm is "TM_CCOEFF_NORMED" method which determines the locations using "normalized cross correlation" The special markers are pasted on specimens to be sure which is template image on source image, and the specimens are subjected to experiments and video recording is taken. Then the template image in the digital image is determined. Every frames in video recording has noisy datas. To eliminate noisy data on the original signal, Gaussian filter is used on each frame and template image. Markers can be found with a good threshold value. The threshold value can be changed between 0 to 1 according to environmental light conditions. Coordinate transformations of the template image along all the frames are followed. But, the coordinate of a target on a frame is generally found as multiple points that are close to each other. Therefore, to reduce extra coordinates and to provide optimal coordinate for a target on frame, arithmetic mean of all points is taken and located to true range. After the displacement values of the targets are found, speed and acceleration datas of the targets have been reached through the derivative operators. The frequency and amplitude values of the structural system can be found by using "Fast Fourier Transform" operator, since the acceleration data is found. In the fourth chapter, different experimental setups applied to the algorithm are mentioned. Also, the graphical comparisions between conventional methods and developed method are given. The applicability of the method used in this study due to the datas obtained from different experiments is also shown in this chapter. In the last section, the results obtained from all the experiments and studies are explained, and suggestions are made for users.