Study Of Key Techniques On Small Field Of View Binocular Vision System

A small field of view binocular vision system (SFVBVS) is proposed for hand-held measuring objects on spot. The perfect measuring range of SFVBVS is 200mm～400/mm as small field of view of cameras. Several key techniques are studied, including the structure, modeling, calibration, binocular matching,3D rebuilding and scanning measurement of marked points. The proposed algorithms are verified to be with high stability, accuracy and use value.The mainly research contents and innovation of this article are as followings:1. Design the structure of the system on the basis of system modeling, analyzing the principle of structure designing.2. Implement the calibration of SFVBVS. The interference of each distortion parameter on image fidelity is analyzed based on six distortion model. A calibration optimal algorithm is proposed. The first step of the algorithm is to obtain initial calibration parameters using OpenCV calibration algorithm. Second step is to optimize six distortion parameters utilize line algorithm. Last step is to optimize the rest parameters using LM algorithm, based on the above distortion parameters and principle points, which the objective function is according to back projection function. In order to improve the stability and precision of binocular calibration based on 2D plane target, three properties are proposed to regulate calibration poses based on homography matrix. Then a strategy to select poses of binocular calibration called five poses set calibration is presented.3. Accomplishing of binocular matching. An algorithm based on similar image geometric feature is proposed. First, the initial matching points which locate in the same or different epipolar lines are identified respectively using epipolar geometric constraint algorithm. According to the position similarity of matching points, maximum vector angle criterion and maximum angle difference criterion are derived. In order to match the points locating in different epipolar lines, utilizing disparity gradient limited constraint as well as maximum vector angle criterion, a mismatching elimination algorithm based on updating strategy is proposed, which can reduce the rate of mismatching and erroneous elimination. For the initial matching points locating in the same epipolar line, maximum angle difference and maximum vector angle criterion are used to extract candidate matching points.4. Scanned measurement of marked points which is stuck on the surface of objects. Through binocular calibration, marked points extraction, binocular matching and 3D building, coordinate of points in the SFVBVS coordinate system is obtained. Reference coordinate system is first established, and then matching of marked points in different coordinate system using distance network is accomplished. In the meantime, transformation matrix can be resolved by the method of adding new points to improve the accuracy of transform 3D points into reference coordinate system. The points which already exist in the reference coordinate system are updated while new points are added into the reference coordinate system. Finally taking average distance as standard, all points in the reference coordinate system are optimized by use of LM method, which will improve the measurement accuracy.