Residual error mitigation in multiview calibration
Multiview calibration is essential for accurate three-dimensional computation. However, multiview calibration cannot be accurate enough because of the tolerances required in some of the intrinsic and extrinsic parameters that are associated with the calibration process, along with fundamental imperfections that are associated with the manufacturing and assembly process itself. As a result, residual error in calibration is left over, with no known methods to mitigate such errors. Residual error mitigation of the invention may be performed inline with a given calibration approach, or as a secondary processing step. Residual error mitigation aims at modifying the original parameters that have been estimated during an initial calibration process. These new, modified parameters are then used for triangulation and depth estimation of scene information. This approach also resolves parameter tolerances that are either too cumbersome to measure, or otherwise impossible to measure for practical stereo and multiview camera production and calibration applications.
1 . A method for error mitigation, comprising:
capturing one or more images viewing an object of known dimensions;
computing one or more object dimensions using an initial set of calibration parameters;
computing an error in the computed one or more dimensions by comparing the computed dimensions to the known dimensions;
determining whether the error has been minimized;
if it is determined that the error has not been minimized, adjusting the initial set of calibration parameters; and
returning to the step of computing the dimensions of the object using an adjusted current set of calibration parameters in place of the initial set of calibration parameters.
2 . The method of claim 1 , wherein the initial set of calibration parameters are is determined according to the steps of:
capturing one or more images viewing a known calibration target having known dimensions at a plurality of distances;
detecting one or more known features of the calibration target in each of the one or more images captured at the plurality of distances; and
obtaining an initial set of calibration parameters by minimizing reprojection error.
3 . The method of claim 2 , wherein the reprojection error is minimized employing a Levenberg-Marquardt process.
4 . The method of claim 1 , wherein the calibration parameters are adjusted in accordance with a slope for the dimensioning error and a learning rate.
5 . The method of claim 1 , wherein the calculated error is related to a yaw error of the object.
6 . The method of claim 1 , wherein the calculated error is related an optical axis alignment in X error of the object.
7 . The method of claim 1 , wherein the calculated error is related to one or more parameters selected from the group of X translation parameter sensitivity of a calibration system, Y translation parameter sensitivity of a calibration system, Z translation parameter sensitivity of a calibration system, pitch angle rotation parameter sensitivity of a calibration system, yaw angle rotation parameter sensitivity of a calibration system, roll angle rotation parameter sensitivity of a calibration system, X optical center parameter sensitivity of a calibration system, and Y optical center parameter sensitivity of a calibration system.
8 . A method for error mitigation with a single calibration target, comprising:
capturing one or more images viewing a known calibration target having known dimensions at a plurality of distances;
detecting one or more known features of the calibration target in each of the one or more images captured at the plurality of distances;
obtaining an initial set of calibration parameters by minimizing reprojection error;
reconstructing one or more locations of a plurality of calibration features using the current set of calibration parameters;
computing the distance between a plurality of combinations of features in the same image spanning a minimum distance;
computing one or more errors in the computed distances using measurements from a predetermined position relative to the camera as the ground truth;
determining whether one or more convergence criteria have been met;
if it is determined that the one or more convergence criteria have not been met, adjusting the initial set of calibration parameters; and
returning to the step of reconstructing one or more locations of a plurality of calibration features using the current set of calibration parameters in place of the initial set of calibration parameters.
9 . The method of claim 8 , wherein the calculated error is related to a yaw error of the object.
10 . The method of claim 8 , wherein the calculated error is related an optical axis alignment in X error of the object.
11 . The method of claim 8 , wherein the calculated error is related to one or more parameters selected from the group of X translation parameter sensitivity of a calibration system, Y translation parameter sensitivity of a calibration system, Z translation parameter sensitivity of a calibration system, pitch angle rotation parameter sensitivity of a calibration system, yaw angle rotation parameter sensitivity of a calibration system, roll angle rotation parameter sensitivity of a calibration system, X optical center parameter sensitivity of a calibration system, and Y optical center parameter sensitivity of a calibration system.
12 . A method for error mitigation, comprising:
capturing one or more images viewing a known calibration target having known dimensions at a plurality of distances;
detecting one or more known features of the calibration target in each of the one or more images captured at the plurality of distances;
obtaining an initial set of calibration parameters by minimizing reprojection error;
reconstructing one or more locations of a plurality of calibration features using the current set of calibration parameters;
computing the distance between a plurality of combinations of features spanning images from multiple depths comprising a virtual calibration target;
computing a standard deviation of each unique measurement of the virtual calibration target;
determining whether the standard deviation has been minimized;
if it is determined that the standard deviation has been minimized, adjusting the initial calibration parameters; and
returning to the step of reconstructing one or more locations of a plurality of calibration features using the current set of calibration parameters in place of the initial set of calibration parameters.
13 . The method of claim 12 , wherein the standard deviation is related to a yaw error of the object.
14 . The method of claim 12 , wherein the standard deviation is related an optical axis alignment in X error of the object.
15 . The method of claim 12 , wherein the standard deviation is related to one or more parameters selected from the group of X translation parameter sensitivity of a calibration system, Y translation parameter sensitivity of a calibration system, Z translation parameter sensitivity of a calibration system, pitch angle rotation parameter sensitivity of a calibration system, yaw angle rotation parameter sensitivity of a calibration system, roll angle rotation parameter sensitivity of a calibration system, X optical center parameter sensitivity of a calibration system, and Y optical center parameter sensitivity of a calibration system.