Research On Bridge Crack Inspection Based On Multi-legged Wall-Climbing Robot Platform

As transport hub of roads, highways, and other transport systems, bridges and their healthstatus related to the lives and safety of the hundreds of millions of the masses. Bridge crackinspection is one of the key content of the bridge safety inspection work. The traditionalbridge crack inspection or recognition technology are difficult to achieve accurate andcomprehensive crack detection, therefore, the research and system development of intelligentbridge cracks recognition algorithms based on computer vision has become a hot topic.Because of the existing interference of a variety of texture and noise on the bridge face, andthe environmental constraints for shooting, crack inspection algorithms with universaladaptability has become the research focus and the difficulty. Furthermore, in order toovercome the shortcomings of the traditional methods, which are hindering the bridge trafficor having difficulty in close-up shots, this paper also proposed a solution using a hexapodrobot adsorbing and climbing on the bridge sidewall and bottom, to achieve a continuousbridge image acquisition. The main research content and results of this paper are described asfollows:1. In the aspect of bridge image preprocessing, common image enhancement methodsare introduced, including the gray stretching and histogram equalization method. After testingthe effect of these methods, it denies their versatility for their drawbacks. Then, it verifies theeffect of median filtering on smooth filtering for images of this paper. Afterward, it proposesthe method with the combination of frequency filtering and spatial filtering, to carry out theimage noise reduction filtering. Here, the focus of discussion is the parameter selection for theFourier transform filtering enhancement.2. In the aspect of the feature extraction of bridge cracks,2-D Discrete WaveletTransform used for the analysis of image detail is introduced. Also, the selection of waveletbasis is studied. Moreover, features in Radon domain are extracted by the Radon transform onthe wavelet coefficients, and used to distinguish the existence of the cracks combined withother wavelet features. The effectiveness of all the features has been verified by theexperimental data.3. The overall effectiveness and accuracy of the proposed features is verified withapplication of artificial neural network pattern recognition algorithm. Experimental data isgiven for the analysis of neural network parameter settings and the best accuracy.4. Hardware, software and control system architecture of the hexapod climbing robot isgenerally described. A linear crawl gait is briefly described, and an image acquisition control method based on this gait is discussed.