Camera Calibration/Parameters help - distance/pixel ratio?

Hi,

I am trying to calculate a distance per pixel ratio for a single camera in matlab. The camera is to face a planar surface, perpendicular to the camera, however the distance between the camera and the surface is likely to change dependent on the setup.

I intent to use the standard Matlab checkerboard as a means of "calibrating" (calculating the distance/pixel ratio, and have thus far computed the camera extrinsics and located the checkerboard points. the distance/pixel is required so that I can estimate the speed of an object tracked through the frame at this planar distance. Is there an easy way of extracting this from the extrinsics or a way of getting matlab to give me the distance between each corner detected and its nearest neighbours so that I can calculate this by hand?

Thank you in advance!

if true
close all; clear all; clc
%
% load camera intrinsics (cameraParams)
load newCameraParams
%
% load image 
imOrig = imread((fullfile('C:\Users\...\Documents\MATLAB\check0003.tif')));
%
% Undistort image, calc. new origin
[im, newOrigin] = undistortImage(imOrig, newCameraParams, 'OutputView', 'full');
%
% Compute Extrinsics for camera
% Detect the checkerboard (reference points) in image
[imagePoints, boardSize] = detectCheckerboardPoints(im);
undistortedPoints = undistortPoints(imagePoints, newCameraParams);
%
% Generate the world coordinates of the checkerboard corners in the
% pattern-centric coordinate system, with the upper-left corner at (0,0).
squareSize = 18.25; % in millimeters
worldPoints = generateCheckerboardPoints(boardSize, squareSize); 
%
% Compute rotation and translation of the camera relative to checkerboard
[R, t] = extrinsics(imagePoints, worldPoints, newCameraParams);
%
% plot figure of corrected image & corner points
figure; imshow(im); title('Corrected lens distortion & detected corners'); 
hold on
plot(imagePoints(:,1), imagePoints(:,2), '+r');
axis('on')
hold off
end

Answer 1

[Dima Lisin]

Hi Yale,

You can actually do better than a distance/pixel ratio. If you know the extrinsics between the camera and the plane, then you can map a point in the image to a point on that plane using the pointsToWorld method of cameraParameters. See this example.

This way the plane does not even have to be perpendicular to the camera's axis.

[Yale Brewer]

Hi,

I had discovered this, but was slightly wary of the computation time required as I am looking to be tracking objects at around 90fps... I suppose I had better give this a go thanks!

[Dima Lisin]

Well, pointsToWorld boils down to matrix multiplication, which MATLAB is very good at. I would not expect that to be a bottleneck, and I would be more worried about the speed of detection and tracking.