Research On Large Scale Vision Measurement Technique Based On Multiple View Geometry Adjustment

Traditional vision measurement system cannot balance the measurement space with the measurement accuracy because of the restriction resolution and view field of the camera. A large scale vision measurement technique in field condition is presented based on multiple view geometry adjustment to solve the problem above. The 3D coordinates of the targets (including coded targets and uncoded targets) distributed on the object to be measured are calculated by taking images using a handheld digital camera from different orientation. The measurement technique is convenient and flexible in operation, and it has high degree of automation. Furthermore, it can maintain the overall measurement accuracy in a large measurement space.Principle analysis and experiments verification to the measurement technique are conducted in this paper, the work of this paper is arranged as follows:1. The mathematical model of the digital camera is established. The projection process of the targets is analyzed. The mathematical relationship between the spatial position and the projection in image plane of the targets is established, which includes the linear model of the perspective projection and the nonlinear model of the lens distortion.2. The extraction of the targets and the recognition of the coded targets are accomplished. The efficient extraction of the targets is realized using a proposed 6-points algorithm according to their imaging features. The robust recognition of the coded targets is accomplished through a presented ellipse-incircle-division algorithm based on their invariance to the affine transformation.3. The transformation relationship between each view is established. The epipolar geometry principle is analyzed and the transformation relationship between the two views that share the maximum common coded targets in number are calculated by the estimation and decomposition of the essential matrix E. Subsequently,3D reconstruction is conducted to the common targets and the result is used as orientation reference. The transformation relationship between the other views and the orientation reference is calculated by linear method. Then, the 3D coordinates of the existing coded targets in the orientation reference are updated and those of the new coded targets are included in the orientation reference by multiple view 3D reconstruction until the transformation relationship between each view is established and the 3D coordinates of all the coded targets are calculated.4. The stereo matching of the uncoded targets is realized. The stereo matching of the uncoded targets is realized by multiple view epipolar constraint on the basis of the known transformation relationship between each view. Subsequently, the 3D coordinates of all the uncoded targets are calculated using multiple view 3D reconstruction.5. The parameters optimization of the measurement system is completed. The nominal values of the digital camera are set as the initial values of the intrinsic parameters when optimization. The initial values of the digital camera intrinsic parameters together with the calculated transformation relationship between each view and 3D coordinates of all the targets are regarded as the parameters to be optimized. Levenberg-Marquardt (LM) optimization algorithm is used for global optimization and the final results of all the parameters are acquired.