Research On The Bridge Crack Detection Method Based On Image Analysis

The bridges are very important components in the transportation system. In recent years, the number of bridges in need of maintenance increases. The demand for bridge safety inspection, particularly bridge crack detection, sharply rises. Traditional bridge crack detection methods are of high cost and high risk. Therefore, it is necessary to propose a new visual crack detection method to complete bridge health evaluation.In this paper, the robotic technology and the digital image processing technology are combined. First, we develop a bionic climbing robot which can climb on rough surface due to its smart structure and bionic design. After loading simple camera, wall images can be acquired wirelessly in real time. Second, a series of image processing algorithms are proposed to detect bridge cracks based on this device. Third, the crack classification is completed by using the classification method based on machine learning. Several aspects of work are introduced as follows:1) Remove the motion blur of the acquired image. In the actual work environment, there is relative movement between the climbing robot and the bridge, thus the generated images have motion blur. Finally, Wiener filter is chosen to restore the image after compared with Constrained Least Squares filter and inverse filter.2) Detect the crack image. First, to extract the crack target, wavelet transform is used to enhance fracture of the crack in the image. Second, to complete the crack image recognition, the surface morphology analysis is applied to extract crack fragments.3) Connect the crack fragments. Due to the above crack detection methods may destroy the connectivity of the cracks. In order to get a more complete and continuous crack, a crack connection method based on KD-tree is proposed. First, obtain the minimum convex polygon of the crack target, and identify the start and end point of each crack fragment. Second, connect the two endpoints from different fragments if their pixel distance is less than the set threshold. Third, fill the connected line segments by testing the grayscale characteristics of the connection region. Finally, a complete crack target is extracted. And the experimental results show that our crack connection method is better.4) Classify the cracks. Due to different shapes of cracks have different degree of harm on the bridge, it is necessary to classify cracks after the complete crack target extracted. In this paper, support vector machine method is used to classify 453 crack pictures based on a series of basic visual characteristics and geometric features. First, six crack characteristics are adopted. Second, train the SVM decision tree classifier based on the training samples. Third, classify the validation sample set by trained classifier. The results show that our classification method has higher accuracy.Currently, the use of digital image processing technology on concrete bridge crack detection has win a widespread concern. This visual detection method not only can free humans from heavy and dangerous work, but also can greatly improve the working efficiency and reduce the cost. It can effectively eliminate the subjective interference. This visual detection method has high practical value and application prospects.