Changes between Version 119 and Version 120 of UvmatHelp


Ignore:
Timestamp:
Dec 7, 2014, 11:34:32 AM (6 years ago)
Author:
sommeria
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • UvmatHelp

    v119 v120  
    566566-'''linear''': general linear transform, including translation and rotation (but no projection effects)
    567567
    568 -'''3D projection:''' this transforms handles projection effects, as needed for stereoscopic view, as well as quadratic distortion, according to the pinhole camera  model ( R.Y. Tsai, 'An Efficient and Accurate Camera Calibration Technique for 3D Machine Vision'. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Miami Beach, FL, pp. 364-374, 1986). We use a more recent version, with the toolbox [->http://www.vision.caltech.edu/bouguetj/calib_doc/] . To use this option, the number of calibration points should be large enough and well spread over the whole image.
    569 
    570 To deduce the 3D physical coordinates from the 2D image coordinates, the ''z'' position, or a plane of cut, defined generally by a laser sheet, must be given as part of the calibration parameters. In UVMAT, it is possible to introduce a set of planes obtained by volume scanning.
    571 
    572 The transform coefficients for each image series are stored under the tag <!GeometryCalib> in the corresponding XML documentation file <!ImaDoc>, described in [#a3.5Imagedocumentationfiles.xml section 3.5].  Calibration coefficients  can be obtained and displayed with the GUI '''geometry_calib.fig'''.
     568-'''3D_linear:''' this transform handles projection effects, needed if the observed plane is not perpendicular to the line of sight. It involves a 3D calibration needed to account for the depth effects occuring in volume scan.
     569
     570-'''3D_quadr:''' this is like 3D_linear, but takes also into account a quadratic deformation by the optics which occurs for wide fields of view (small focal lengths).
     571
     572-'''3D_extrinsic:''' this is like 3D_quadr, but uses intrinsic parameters of the camera, as explained below. ''''''
     573
     574The 3D options involve a full 3D calibration relying on the [attachment:3D_view.pdf pinhole camera  model]. The method was first proposed by R.Y. Tsai, 'An Efficient and Accurate Camera Calibration Technique for 3D Machine Vision'. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Miami Beach, FL, pp. 364-374, 1986). We use a more recent version, with the toolbox [->http://www.vision.caltech.edu/bouguetj/calib_doc/] . 3D calibrations are done in two steps. The camera'' intrinsic parameters'', which are the focal length and the quadratic deformation coefficient, are first determined by different views of the same grid observed at different angles. Then the ''extrinsic parameters'', which represent the rotation angles and translation of the physical coordinates with respect to the camera, are obtained with a single image of the grid positioned in a known plane $z=cte$. The option 3D_extrinsic allows the user to do only the second step from known intrinsic parameters. Those depend only on the camera with its objective lens and focus adjustement. Note that these 3D options require a calibration grid, with a sufficient number of calibration points covering the whole image.
     575
     576The transform coefficients for each image series are stored in the corresponding XML documentation file <!ImaDoc>, described in [https://servforge.legi.grenoble-inp.fr/#a3.5Imagedocumentationfiles.xml section 3.5],  under the tag <!GeometryCalib>.   Calibration coefficients  can be displayed with the GUI '''geometry_calib.fig''' described below. Field transform from pixel to physical parameters is performed by the function ''phys.m'' in UVMAT/transform_field, which calls the pointwise transform functions  ''phys_XYZ.m'' (from image to physical coordinates) and ''px_XYZ.m'' (from physical to image coordinates), as well as ''phys_ima.m'' which transforms images.  When an image or PIV velocity field is opened by uvmat, the transform function 'phys' is automatically loaded as described in section 4.6.
    573577
    574578=== 8.2 The GUI geometry_calib.fig ===
     
    595599Alternatively the command '''[REPLICATE]''' can be used to calibrate a whole set of experiments in a 'Campaign', using an overview of the data set provided by the GUI '''data_browser.fig''', described in [#a3.7Dataorganisationinaproject section 3.7].
    596600
    597 -'''3D calibration''': 3D projection is handled by the options in '''[calib_type]''' '3D_lin' or '3D_quad' ((if non-linear distortion is significant). By default, the set of calibration points is assumed to be contained in a single plane ''z''=0. For a correct determination of the 3D features, the normal to this plane must be tilted with respect to the line of view. Otherwise this problem of indetermination can be resolved by using a set of (typically 5-10) calibrations images using a plane grid with different tilting angles (for precision the grid must cover a large area of the view field). On each image, get calibration points with the tool '''[!Tools/Detect grid]''', introducing the appropriate grid mesh. Do not fill info on ''z'' coordinates. Store the points each time (without applying calibration at this stage), which fills the list [ListCoordFiles] of file names. Then introduce a last grid image which will be considered as defining the orientation of the ''z'' axis, perpendicular to the grid. Detect points on this last image, but instead of storing them, apply the calibration with the option 3D_linear or 3D_quadr . For calibration performed inside water, the option of introducing the ''z'' position of water upper surface is provided.
     601-'''3D calibration''': 3D projection is handled by the options in '''[calib_type]''' '3D_lin' or '3D_quad' ((if non-linear distortion is significant). By default, the set of calibration points is assumed to be contained in a single plane ''z''=0. For a correct determination of the 3D features, the normal to this plane must be tilted with respect to the line of view. Otherwise this problem of indetermination can be resolved by using a set of (typically 5-10) calibrations images using a plane grid with different tilting angles (for precision the grid must cover a large area of the view field). On each image, get calibration points with the tool '''[!Tools/Detect grid]''', introducing the appropriate grid mesh. Do not fill info on ''z'' coordinates. Store the points each time (without applying calibration at this stage), which fills the list [ListCoordFiles] of file names. Then introduce a last grid image which will be considered as defining the orientation of the ''z'' axis, perpendicular to the grid. Detect points on this last image, but instead of storing them, apply the calibration with the option 3D_linear or 3D_quadr. A non-zero ''z'' position of this grid can be introduced by a z translation performed with '''Tools/Translate points'''.
    598602
    599603-'''Intrinsic parameters''': the previous procedure first determines the extrinsic parameters which characterize the camera optics (focal lengths and nonlinear deformation parameter). Then the extrinsic parameters, translation and rotation of the camera with respect to the reference grid, are determined on the last grid image. if the same optics is used in a new experiment, it is possible to skip the multiplane detection, importing the intrinsic parameters from a previous <!ImaDoc> file by the menu bar tool '''[!Import/Intrinsic]''' parameters, then applying the calibration with the option '3D_extrinsic' with the reference grid image only.
    600604
    601 -'''Section planes:''' deducing the physical coordinates from image coordinates requires information on the section plane. The default assumption is that the objects in the image are in the plane ''z''=0, but UVMAT can handle volume scanning by a laser plane. A set of section planes can be defined by their origin positions and rotation angle vectors. Theses planes are labelled by a ''z index'', assumed to be the frame index j (case of volume scan), or the index i modulo the number of slices !NbSlice (case of multiplane scan). The chosen option can be doucumented in the file <!ImaDoc> by the menu bar command '''[Tools/set planes]''' of '''geometry_calib.fig'''. A dialog box appears for entering the set of section plane positions ''z'', as lower value, upper value and increment. Similarly an angular scan be be introduced. After introduction of the multi-plane information, the z-index is displayed in the frame '''[FileIndices]''' of '''uvmat.fig'''. The local z position of the mouse pointer, assuemed to lay on the current section plane, is then displayed in '''[text_display]'''.
     605-'''Section planes:''' deducing the physical coordinates from image coordinates requires information on the section plane. The default assumption is that the objects in the image are in the plane used for calibration, but UVMAT can handle volume scanning by a laser plane. A set of section planes can be defined by their origin positions and rotation angle vectors. Theses planes are labelled by a ''z index'', assumed to be the frame index j (case of volume scan), or the index i modulo the number of slices !NbSlice (case of multiplane scan). The chosen option can be documented in the file <!ImaDoc> by the menu bar command '''[Tools/set slice]''' of '''uvmat.fig'''. A dialog box appears for entering the set of section plane positions ''z'', as lower value, upper value and increment. An angular scan of the laser sheet can be also introduced. After introduction of the multi-plane information, the z-index is displayed in the frame '''[FileIndices]''' of '''uvmat.fig'''. The local z position of the mouse pointer, assumed to lay on the current section plane, is then displayed in '''[text_display]'''.
     606
     607-'''Water surface''': calibration is best performed inside water to account for refraction effects. However it may be easier to perform it in air, and then take into account the refraction effect. For that purpose the water surface position z (assumed normal to the z coordinate) and refraction index can be introduced with  '''[Tools/set slice].'''
    602608
    603609=== 8.3 Structure of the XML file ===