Surface Defect3D Reconstruction And Inspection Evaluation Method Based On Binocular Stereo Vision

Aimed at the inspection requirement of structural surface flaw detection in civil engineering field, the binocular stereo vision model, camera calibration methed and the3D reconstruction algorithm are studied in paper. The surface3D reconstruction and evaluation system is developed based on these algorithms, and the inspection experiments are conducted by the integrated system in paper.The main content are introduced as follows:Firstly, the mathematical model of binocular stereo vision and camera calibration method are analyzed and discussed respectively. Compared Zhang Zhengyou calibration method with the method of DLT and Tsai’s two-step through experiment, the result indicates that Zhang’s calibration method is the most accurate. On the base of Zhang’s2D plane targets calibration method, the image distortion is also corrected. The image pair in the parallel binocular stereo vision model are obtained by epipolar rectification. The image pair in the parallel binocular stereo vision model are obtained by epipolar rectification. The image pre-processing method is used to solve the denoising and brightness differences problems of the corrected image pair. The normalized cross correlation coefficient is adopted to get the initial parallax by region matching algorithm, and the final parallax is optimized by tha maximum and minimu parallax method based on probability distribution histogram proposed in paper. The three-dimensional world coordinate of surface meassurment points are calculated according to the parallel binocular stereo vision model, the free surface can be generated by Delaunay triangulation on the base of3D points cloud, accordingly the surface defect section curvesis obtained.Secondly, to overcome the shortcomging of feature extraction which couldn’t extract any surface point, a semi-automatic point selection surface defect inspection algorithm is proposed, including the computation of defect’s length, width, depth and area. The structural surface defect3D reconstruction and evaluation system is integrated on MATLAB platform on the base of upper algorimthm. The precision experiment of proposed algorithm and comparative trial with manual point extraction method are all carried with the system, the results show that the proposed method in paper has characteristic of simple operation and high precision.Thirdly, the inspection and construction experiments are conducted to detect the concrete cylinder, the concrete spalling and preiection, and the distorted aluminum foam plate by the integrated system in paper, the experiment results show that the system has more accuracy in detecting the length and width and area than detecting depth, and the reconstructed surface can correctly reflect the actual shape of the defects.