What is the rigth use of MATLAB built-in functions for visual odometry?

2 views (last 30 days)
I am working on a visual odometry code in MATLAB. I am using the following example ( estimateEssentialMatrix ) to obtain the essential matrix. You can open this example typing:
openExample('vision/EstimateEssentialMatrixFromAPairOfImagesExample')
to the Command Window. Then, I used
[relativeOrientation,relativeLocation] = relativeCameraPose(E,cameraParams,inlierPoints1,inlierPoints2);
[rotationMatrix,translationVector] = cameraPoseToExtrinsics(relativeOrientation,relativeLocation);
to recover the rotational matrices and translation vector. Then, I concatenated and plotted the translation vector (which indicates the location of the camera),
T_t = T_t + R_t * translationVector';
R_t = R_t * rotationMatrix';
location = vertcat(location,[T_t(1),T_t(3)]);
plot3(location(:,1),zeros(size(location(:,1),1),1), location(:,2))
where initially,
R_t = eye(3);
T_t = [0;0;0];
location = [0,0];
But, I am not getting the right result. I presume the issue is with the cameraParams object that contains the parameters of the camera.
I used a function ( ReadCameraModel) that was provided along with the data-set to obtain the camera intrinsics and undistortion LUT. The i/o of the function are as follows:
% ReadCameraModel - load camera intrisics and undistortion LUT from disk
%
% [fx, fy, cx, cy, G_camera_image, LUT] = ReadCameraModel(image_dir, models_dir)
%
% INPUTS:
% image_dir: directory containing images for which camera model is required
% models_dir: directory containing camera models
%
% OUTPUTS:
% fx: horizontal focal length in pixels
% fy: vertical focal length in pixels
% cx: horizontal principal point in pixels
% cy: vertical principal point in pixels
% G_camera_image: transform that maps from image coordinates to the base frame of the camera.
% For monocular cameras, this is simply a rotation.
% For stereo camera, this is a rotation and a translation to the left-most lense.
% LUT: undistortion lookup table. For an image of size w x h, LUT will be an
% array of size [w x h, 2], with a (u,v) pair for each pixel. Maps pixels
% in the undistorted image to pixels in the distorted image
I used the following code to get the cameraParams object:
[fx, fy, cx, cy, G_camera_image, LUT] = ReadCameraModel('./stereo/centre','./model');
K = [fx 0 cx;0 fy cy;0 0 1]; %Intrinsic Matrix of the camera
cameraParams = cameraParameters('IntrinsicMatrix',K);
Is this the right way of getting the cameraParams object? If yes, what is it that I am doing wrong?

Answers (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!