Research On The Key Techniques Of Quad-nocular Stereo Measurement

3D profile measuring technique is highly demanded and widely used in aerospace manufacturing,machinery industries, automobile making, shipbuilding, die manufacturing and biomedicine etc.Based on the technique of binocular stereo measurement, the scheme of quad-nocular stereomeasurement is proposed, which enlarges the single-pass measurement range of measurement systemand automatic registration of measurement data in different visual angles is realized. Main contentsand contributions are as follows.The calibration of quad-nocular stereo measurement system based on a calibration cross isdeeply researched. The calibration cross is composed of two detachable rulers, on which only twodistances are required to be known in advance. The target points are not required to be coplanar andthe positions of target points are not required to be accurately measured, so the calibration cross isconvenient for manufacture and flexible for on-site calibration. Target points are designed to be withthe unique IDs which make the match of two pictures and the match of image points and3D pointsreliable and stable. The intrinsic parameters of the four cameras are calculated precisely and the3Dtarget point sets are reconstructed in each camera coordinate system respectively by using the multipleview geometry constraint and a generic sparse bundle adjustment. Then the extrinsic parameters arecalculated by using the rigid transform of the two3D point sets. This method is simple and flexiblefor field operation because it only needs the calibration cross to be photographed by four camerassimultaneously once and then separately several times. The accuracy of the proposed calibrationmethod is verified by experiments.A automatic registration of3D points clouds based on quad-nocular measurement is proposed. Inthe quad-nocular stereo measurement system, dense point correspondences are acquired by combiningthe feature matching and area matching methods and3D points of object’s surface profile arereconstructed by triangulation.3D points clouds reconstructed from each two images successively arelocated in different camera coordinate systems and registration is required to convert them to the samecamera coordinate system. The rotate matrix and translation vector required in the registration processare precisely calculated in the process of system calibration, thus automatic registration of3D pointclouds is realized and the accuracy of which is verified by experiments.The algorithm of data integration is studied and among which the detection of overlapping areasis the most important step. The method of projection is adopted to detect overlapping areas and the projection matrix can be precisely calculated in the quad-nocular stereo measurement, thus theoverlapping areas can be detected precisely. Regular triangular mesh model is established according tothe correspondences of3D point and image point before integration, then one mesh model is definedas the benchmark. The overlapping areas between the benchmark and other mesh models are detectedand adjusted optimally. Then the redundancy is merged and removed, and the gaps are connectedsmoothly. Finally, a complete topological mesh surface can be generated.