The Research On Three-dimensional Reconstruction Algorithm Of Material Stack Based On Multi-vision

It is a very important part of the logistics information management for many large engineering enterprises to realize the effective measurement in the process of the material stack storage and inventory. The three-dimensional reconstruction algorithm of material stack based on Multi-vision (tri-vision) is researched in this thesis, basing on the analysis of a variety of existing three-dimensional measurement methods.First, the experimental device is designed successfully in this subject. Artificial feature points on the surface of the material stack are made through the black and white grid projection, since the material stack itself has no feature points. Considering the radial distortion of camera, the camera calibration is completed by using the Tsai two-step method which is based on the stereo calibration module in this thesis. This method is suitable for the volume measurement of the actual material stack pile, since it has no strict requirements on the calibration target.The original images of the material stack are preprocessed in this thesis to get the high SNR images which can reflect the actual shape information of the material stack. The preprocessing mainly include:graying of color image, histogram equalization, de-noising with smoothing filtering, sharpening with homomorphic filter. The enhanced material stack images are dealt with to get the corresponding binary imaged, and then the binary imaged are edge detected on to get the edge of the black and white grid on the material stack surface.The algorithm of increasing the feature points on the material stack surface is studied to improve the precision of three-dimensional reconstruction in this thesis, that is designed on the basis of artificially completed feature points extraction and matching on the material stack surface. Firstly, the middle points of the image edge are matched through the distance constraints of the matched feature points. Then, the other points of the image edge are matched by using edge matching algorithm guided by matched points and with the edge sequential consistency constraints. Eventually, the algorithm realizes the number increasing of the matched points on the material stack surface.Finally, according to the matching information, the three-dimensional reconstruction and tilt correction of the feature points are completed in thesis. Experiments are done to measure the volume of several different material stacks before and after increasing matched feature points on the material stack surface under the different sizes of the grid projection. Results showing that the increasing feature points algorithm can effectively reduce the volume measurement error. This is very meaningful for the reason that it can greatly reduce the artificial matching workload of the feature points when the algorithm is used in the volume measurement of actual large material stack.To a certain extent, the automatic volume measurement of large material stack can be realized in this thesis. It proves the feasibility and accuracy of this method through simulation experiments in laboratory. This method is simple, and its cost is low.