source: trunk/src/toolbox_calib/init_intrinsic_param.m @ 908

%=======================================================================
% Copyright 2008-2015, LEGI UMR 5519 / CNRS UJF G-INP, Grenoble, France
%   http://www.legi.grenoble-inp.fr
%   Joel.Sommeria - Joel.Sommeria (A) legi.cnrs.fr
%
%     This file is part of the toolbox UVMAT.
%
%     UVMAT is free software; you can redistribute it and/or modify
%     it under the terms of the GNU General Public License as published
%     by the Free Software Foundation; either version 2 of the license,
%     or (at your option) any later version.
%
%     UVMAT is distributed in the hope that it will be useful,
%     but WITHOUT ANY WARRANTY; without even the implied warranty of
%     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
%     GNU General Public License (see LICENSE.txt) for more details.
%=======================================================================

%init_intrinsic_param
%
%Initialization of the intrinsic parameters.
%Runs as a script.
%
%INPUT: x_1,x_2,x_3,...: Feature locations on the images
%       X_1,X_2,X_3,...: Corresponding grid coordinates
%
%OUTPUT: fc: Camera focal length
%        cc: Principal point coordinates
%             kc: Distortion coefficients
%        alpha_c: skew coefficient
%        KK: The camera matrix (containing fc, cc and alpha_c)
%
%Method: Computes the planar homographies H_1, H_2, H_3, ... and computes
%        the focal length fc from orthogonal vanishing points constraint.
%        The principal point cc is assumed at the center of the image.
%        Assumes no image distortion (kc = [0;0;0;0])
%
%Note: The row vector active_images consists of zeros and ones. To deactivate an image, set the
%      corresponding entry in the active_images vector to zero.
%
%
%Important function called within that program:
%
%compute_homography.m: Computes the planar homography between points on the grid in 3D, and the image plane.
%
%
%VERY IMPORTANT: This function works only with 2D rigs.
%In the future, a more general function will be there (working with 3D rigs as well).


if ~exist('two_focals_init'),
    two_focals_init = 0;
end;

if ~exist('est_aspect_ratio'),
    est_aspect_ratio = 1;
end;

check_active_images;

if ~exist(['x_' num2str(ind_active(1)) ]),
    click_calib;
end;


fprintf(1,'\nInitialization of the intrinsic parameters - Number of images: %d\n',length(ind_active));


% Initialize the homographies:

for kk = 1:n_ima,
    eval(['x_kk = x_' num2str(kk) ';']);
    eval(['X_kk = X_' num2str(kk) ';']);
    if (isnan(x_kk(1,1))),
        if active_images(kk),
            fprintf(1,'WARNING: Cannot calibrate with image %d. Need to extract grid corners first.\n',kk)
            fprintf(1,'         Set active_images(%d)=1; and run Extract grid corners.\n',kk)
        end;
        active_images(kk) = 0;
    end;
    if active_images(kk),
        eval(['H_' num2str(kk) ' = compute_homography(x_kk,X_kk(1:2,:));']);
    else
        eval(['H_' num2str(kk) ' = NaN*ones(3,3);']);
    end;
end;

check_active_images;

% initial guess for principal point and distortion:

if ~exist('nx'), [ny,nx] = size(I); end;

c_init = [nx;ny]/2 - 0.5; % initialize at the center of the image
k_init = [0;0;0;0;0]; % initialize to zero (no distortion)



% Compute explicitely the focal length using all the (mutually orthogonal) vanishing points
% note: The vanihing points are hidden in the planar collineations H_kk

A = [];
b = [];

% matrix that subtract the principal point:
Sub_cc = [1 0 -c_init(1);0 1 -c_init(2);0 0 1];

for kk=1:n_ima,
    
    if active_images(kk),
        
        eval(['Hkk = H_' num2str(kk) ';']);
        
        Hkk = Sub_cc * Hkk;   
        
        % Extract vanishing points (direct and diagonals):
        
        V_hori_pix = Hkk(:,1);
        V_vert_pix = Hkk(:,2);
        V_diag1_pix = (Hkk(:,1)+Hkk(:,2))/2;
        V_diag2_pix = (Hkk(:,1)-Hkk(:,2))/2;
        
        V_hori_pix = V_hori_pix/norm(V_hori_pix);
        V_vert_pix = V_vert_pix/norm(V_vert_pix);
        V_diag1_pix = V_diag1_pix/norm(V_diag1_pix);
        V_diag2_pix = V_diag2_pix/norm(V_diag2_pix);
        
        a1 = V_hori_pix(1);
        b1 = V_hori_pix(2);
        c1 = V_hori_pix(3);
        
        a2 = V_vert_pix(1);
        b2 = V_vert_pix(2);
        c2 = V_vert_pix(3);
        
        a3 = V_diag1_pix(1);
        b3 = V_diag1_pix(2);
        c3 = V_diag1_pix(3);
        
        a4 = V_diag2_pix(1);
        b4 = V_diag2_pix(2);
        c4 = V_diag2_pix(3);
        
        A_kk = [a1*a2  b1*b2;
            a3*a4  b3*b4];
        
        b_kk = -[c1*c2;c3*c4];
        
        
        A = [A;A_kk];
        b = [b;b_kk];
        
    end;
    
end;


% use all the vanishing points to estimate focal length:


% Select the model for the focal. (solution to Gerd's problem)
if ~two_focals_init
    if b'*(sum(A')') < 0,
        two_focals_init = 1;
    end;
end;

    

if two_focals_init
    % Use a two focals estimate:
    f_init = sqrt(abs(1./(inv(A'*A)*A'*b))); % if using a two-focal model for initial guess
else
    % Use a single focal estimate:
    f_init = sqrt(b'*(sum(A')') / (b'*b)) * ones(2,1); % if single focal length model is used
end;


if ~est_aspect_ratio,
    f_init(1) = (f_init(1)+f_init(2))/2;
    f_init(2) = f_init(1);
end;

alpha_init = 0;

%f_init = sqrt(b'*(sum(A')') / (b'*b)) * ones(2,1); % if single focal length model is used


% Global calibration matrix (initial guess):

KK = [f_init(1) alpha_init*f_init(1) c_init(1);0 f_init(2) c_init(2); 0 0 1];
inv_KK = inv(KK);


cc = c_init;
fc = f_init;
kc = k_init;
alpha_c = alpha_init;


fprintf(1,'\n\nCalibration parameters after initialization:\n\n');
fprintf(1,'Focal Length:          fc = [ %3.5f   %3.5f ]\n',fc);
fprintf(1,'Principal point:       cc = [ %3.5f   %3.5f ]\n',cc);
fprintf(1,'Skew:             alpha_c = [ %3.5f ]   => angle of pixel = %3.5f degrees\n',alpha_c,90 - atan(alpha_c)*180/pi);
fprintf(1,'Distortion:            kc = [ %3.5f   %3.5f   %3.5f   %3.5f   %5.5f ]\n',kc);