Research On Binocular Stereo Vision Methods Based On One-shot Gray Coded Structured Light Pattern

Optical3D measurement techniques have been widely used in many areas, such asindustrial automatic detection, product quality control, reverse engineering, biomedical,virtual reality, anthropometric measurement and so on. But most of the techniques are onlysuited for static objects or scenes. In practice, there are kinds of moving or deforming objectswhich need to be reconstructed or measured. One-shot coded structured light pattern, withinherent advantages for reconstructing moving objects, has good research value andapplication prospect. The practical application of existing one-shot structured light codingtechniques are obstructed, because of the introduction of colors which have difficulties inidentifying colors when measuring non-neutral surfaces, or undue reliance on neighborhoodinformation which can lead to erroneous window decoding on the discontinuities of the objectsurface, or low spatial resolution and accuracy. In this thesis, several key technologies of thebinocular stereo vision based on one-shot structured pattern are deeply researched, includingthe design and correspondence of the structured pattern, the system model and theoptimization algorithm of system parameters. We aim to develop a3D reconstruction methodwith high adaptability, high spatial resolution and accuracy, which is suitable for both staticand moving objects.In view of the weakness of traditional one-shot structured pattern, we designed a newone-shot2D gray-coded pattern, analyzed the uniqueness of pattern features, and developedthe feature extraction algorithm and stereo correspondence algorithm based on dynamicprogramming.Based on the analysis of the relationship between the space points and image points inbinocular stereo vision system, we choose the system model parameters, and proposed a newtriangulation method to compute the3D space points under the minimization of re-projectionerror. In addition, we compared the reconstruction error between the new triangulationmethod and the method based on the midpoint of the common perpendicular.Calibration of system parameters is one of the key technologies of binocular stereovision measurement. The goodness of calibration directly determines the accuracy of thesubsequent3D reconstruction. In this thesis, a set of compact and efficient formulas for thecomputation of Jacobian matrix in the non-linear optimization module is proposed, whichintegrates the chain rule derivation, the incremental representation of rotation matrix, theselection of parameters’ scales and the application of sparse matrix. Experiments areconducted to verify the proposition’s concision, high-speed and high accuracy.Finally, using the proposed binocular stereo vision method based on one-shot two-axisgray coded structured pattern to reconstruct the static and moving objects, experimental results show that the proposed method is suitable for the free-form surface lacking of textures,the discontinuous surface and the non-neutral complex surface, and can obtain the3Dmorphology of moving objects with high spatial resolution and good reconstruction accuracy.