Research And Application Of Binocular Fisheye Based 3D Reconstruction Technology Of Coal

The blockage of raw coal bunkers is a general problem in thermal power industry.The monitoring of material level in the bunkers are meaningful to the timely detection of blockages Existing contact measurement methods have the shortcoming of insufficient accuracy,while non-contact laser ranging measurement requires expensive equipment which makes it difficult to popularize.In addition,the raw coal bunker has the characteristics of large scenes and wide range of material depth changes,and it is difficult to take a panoramic view with normal cameras Therefore,the fisheye stereo photogrammetry scheme for bunker level measurement is studied in the dissertation,focusing on fisheye camera calibration and stereo matching strategies.The main innovative results achieved are as follows(1)A volume measurement system suitable for the raw coal bunker in thermal power plant is designed.To measure large-scale targets at short distances,a binocular fisheye vision system is established.Due to the large range of level change in different working conditions,the setting of the baseline of the binocular system is studied to ensure the accuracy of level measurement at different depths meets the requirement.(2)Given that the findChessboardCorners function in OpenCV fails to detect checkerboard corners in highly distorted fisheye image,a chessboard corner extraction algorithm based on orthogonal line intersection detection is proposed.Experimental results show that the algorithm can extract high-precision control points in fisheye image,and is robust to cases where chessboard is incompletely imaged.This paper studies template-based fisheye camera calibration algorithm,and the camera is calibrated based on the stereographic projection,equidistant projection,and equal-solid angle projection models separately.Results with the smallest reprojection error is selected for subsequent operations.(3)Given the low brightness and weak surface texture characteristics of the coal pile,the matching cost calculation and aggregation stategies of the local stereo matching algorithm are researched to design a local stereo matching algorithm suitable for the coal pile.According to the results of comparative experiments,SIFT descriptors are used for matching costs calculation.Through the comparative analysis study of the guided image filter and its improved versions,a two-stage cascaded guided image filter which uses the edge-preserving filter to process the coefficient matrix is proposed.The filter can achieve improved edge preservation performance.The two-stage cascaded guided image filter is applied for image preprocessing and matching cost aggregation in the volume measurement system.A volume measurement system based on a binocular fisheye vision system,which calculates the volume of the stockpile based on the reconstructed 3D point cloud,is built in this dissertation.Effectiveness of the system is tested based on laboratory data.Experimental results show that the system is stable for 3D reconstruction and the relative error of volume measurement results meet the accuracy requirement within 10%of the relative error.The system can provide important reference and theoretical support for field applications.