Bridge Crack Measurement System Based On Binocular Stereo Vision

Bridge crack measurement is an essential part of bridge project, and computer vision measurement technology has become a hot spot of computer technology at home and abroad as a rising technology, due to its high precision, non-contact characteristic, high efficiency and good real-time performance. Computer vision measurement technology and bridge crack measurement have been integrated in this paper to study the important role and the advantages of binocular stereo vision technology in bridge crack measurement.Small four-rotor unmanned aerial vehicle(UAV) equipped with binocular imaging system was used to take photos, which would been processed with some image processing technologies. Simple introduction For UAV was made and a stability controlment method of the UAV was designed. A variety of pre-processing algorithms were introduced and their advantages and disadvantages were compared. Linear transformation method of gray image was selected to enhance image because of its good performance so that the image grayscale contrast is more obvious. A custom 3*3 filtering de-noising template was choosed to denose the noise of image. The Sobel operator was selected to detect and extract the edge of bridge crack. After these steps, the final bridge crack binary image was got,whose outline and background were split open.The significance of camera calibration was described and several major camera calibration methods was analyses detailedly. The principle and algorithm process of each method was in-depth introduced, and their advantages and disadvantages was compared. Finally, A classic method,Zhang Zhengyou calibration method was selected.This paper focused on the work of the image matching. Two main types of image matching algorithms were described. By analyzing the characteristics of some algorithms with characteristics of binary bridge cracks image was taken into consideration, an image matching method was designed. First of all, the constraint of epipolar and gray value narrowed the candidate points down to a small number of candidate points. Then normalized cross-correlation(NCC) algorithm was used to make a similarity measurement to find the right point ultimately. From Simulations of experiment, this method has been validated that it is good for matching the relative points pair of the bridge cracks.2 sets of photos were taken from bridge in three different angles respectively to test the system performance. Binocular vision system was used to measure the width and the length of cracks, and was compared with the results by monocular vision system under the same conditions.The results show that the binocular vision measurement system can be used in practical well with stronger robustness and higher precision. The software was finished by C++ computer language written on Qt platform.The work we have done was summarised in the last of the article, and an outlook for the future work which need to improve or expand was summarised in here too.