I want to project a point in 3D space into 2D image coordinates. I have the calibrated intrinsics and extrinsics of the camera I'm using. I have the camera matrix K and distortion coefficients D. However, I want the projected image coordinates to be of the undistorted image.
From my research, I found two ways to do this.
Use opencv's getOptimalNewCameraMatrix function to compute a new undistorted image's camera matrix K'. Then use this K' in opencv's projectPoints function, with the distortion parameters set to 0, to get the projected point.
Use projectPoints function using the raw camera matrix K, along with the distortion coefficients D in this function and get the projected point.
Should the output of both methods match?
I think that there is something missing in your thought.
Camera matrix K and dist. coefficent D are the parameters for make the undistortion (if your lens is distorting the image like in a fisheye). They are what is called intrinsic camera parameters.
If we change terms from computer vision to computer graphics, those parameters are the one you use for defining the frustum of the view, and, for example, they are used for getting the focal length of the camera.
But they are not enough to do the projection stuff.
For the projection, if you think in a computer graphics term (like opengl, for instance) you need to have the model-view-projection matrix. The model matrix is the matrix that specifies the position of the object in the world. The view matrix specifies the position of the camera, and the projection matrix specify the frustum (focal angle, perspective distortion, etc).
If you want to know how to transform the points of the model from 3d to 2d (or viceversa) you need the projection and the view matrixes (you have the model matrix because you have the 3d points from which you want to start). And in computer vision the view matrix is called estrinsic parameters.
So, you need the estrinsic parameters too, that are the position of the camera in the world. That is, for instance, those parameters are the rvec and tvec that cv:: projectPoints needs.
If you want to compute them, they are exactly the output of cv::solvePnP that do the opposite of what you want to do: from some known 3d points coupled with the known 2d projection on them on the camera screen, this function gives you the estrinsic parameters (from which you can get the view matrix for some opengl-opencv-augmented-reality-whatever application via cv::Rodrigues).
Last note: while the instrinsic parameters are fixed in all the pictures you shoot with a camera (while you don't change the focal length of course), the estrinisc parameters changes every time you move the camera for take a new picture from a different view point (that is: this changes the perspective point of view, so the 3D-2D projection you want to find)
Hope could help!
Related
I am using opencv to calibrate my webcam. So, what I have done is fixed my webcam to a rig, so that it stays static and I have used a chessboard calibration pattern and moved it in front of the camera and used the detected points to compute the calibration. So, this is as we can find in many opencv examples (https://docs.opencv.org/3.1.0/dc/dbb/tutorial_py_calibration.html)
Now, this gives me the camera intrinsic matrix and a rotation and translation component for mapping each of these chessboard views from the chessboard space to world space.
However, what I am interested in is the global extrinsic matrix i.e. once I have removed the checkerboard, I want to be able to specify a point in the image scene i.e. x, y and its height and it gives me the position in the world space. As far as I understand, I need both the intrinsic and extrinsic matrix for this. How should one proceed to compute the extrinsic matrix from here? Can I use the measurements that I have already gathered from the chessboard calibration step to compute the extrinsic matrix as well?
Let me place some context. Consider the following picture, (from https://docs.opencv.org/2.4/modules/calib3d/doc/camera_calibration_and_3d_reconstruction.html):
The camera has "attached" a rigid reference frame (Xc,Yc,Zc). The intrinsic calibration that you successfully performed allows you to convert a point (Xc,Yc,Zc) into its projection on the image (u,v), and a point (u,v) in the image to a ray in (Xc,Yc,Zc) (you can only get it up to a scaling factor).
In practice, you want to place the camera in an external "world" reference frame, let's call it (X,Y,Z). Then there is a rigid transformation, represented by a rotation matrix, R, and a translation vector T, such that:
|Xc| |X|
|Yc|= R |Y| + T
|Zc| |Z|
That's the extrinsic calibration (which can be written also as a 4x4 matrix, that's what you call the extrinsic matrix).
Now, the answer. To obtain R and T, you can do the following:
Fix your world reference frame, for example the ground can be the (x,y) plane, and choose an origin for it.
Set some points with known coordinates in this reference frame, for example, points in a square grid in the floor.
Take a picture and get the corresponding 2D image coordinates.
Use solvePnP to obtain the rotation and translation, with the following parameters:
objectPoints: the 3D points in the world reference frame.
imagePoints: the corresponding 2D points in the image in the same order as objectPoints.
cameraMatris: the intrinsic matrix you already have.
distCoeffs: the distortion coefficients you already have.
rvec, tvec: these will be the outputs.
useExtrinsicGuess: false
flags: you can use CV_ITERATIVE
Finally, get R from rvec with the Rodrigues function.
You will need at least 3 non-collinear points with corresponding 3D-2D coordinates for solvePnP to work (link), but more is better. To have good quality points, you could print a big chessboard pattern, put it flat in the floor, and use it as a grid. What's important is that the pattern is not too small in the image (the larger, the more stable your calibration will be).
And, very important: for the intrinsic calibration, you used a chess pattern with squares of a certain size, but you told the algorithm (which does kind of solvePnPs for each pattern), that the size of each square is 1. This is not explicit, but is done in line 10 of the sample code, where the grid is built with coordinates 0,1,2,...:
objp[:,:2] = np.mgrid[0:7,0:6].T.reshape(-1,2)
And the scale of the world for the extrinsic calibration must match this, so you have several possibilities:
Use the same scale, for example by using the same grid or by measuring the coordinates of your "world" plane in the same scale. In this case, you "world" won't be at the right scale.
Recommended: redo the intrinsic calibration with the right scale, something like:
objp[:,:2] = (size_of_a_square*np.mgrid[0:7,0:6]).T.reshape(-1,2)
Where size_of_a_square is the real size of a square.
(Haven't done this, but is theoretically possible, do it if you can't do 2) Reuse the intrinsic calibration by scaling fx and fy. This is possible because the camera sees everything up to a scale factor, and the declared size of a square only changes fx and fy (and the T in the pose for each square, but that's another story). If the actual size of a square is L, then replace fx and fy Lfx and Lfy before calling solvePnP.
calibrateCamera() provides rvec, tvec, distCoeff and cameraMatrix whereas solvePnP() takes cameraMatrix, distCoeff as input and provides rvec, tvec as output. What is the difference between these two functions?
cv::calibrateCamera(...)
The function estimates the following parameters of a monocular camera from several views of a calibration pattern. The geometry of this pattern is usually known (i.e. it can be a chessboard):
The linear intrinsic parameters: the focal lengths in terms of pixels (these are basically scale factors), the principal point which would be ideally in the center of the image, and sometimes a skew coefficient between the x and the y axis (but this is often zero).
The non-linear intrinsic parameters: the previously mentioned parameters are forming the linear camera matrix, but there are also some non-linear parameters in the tranformation from the 3D camera to the 2D image plane, i.e. the lens distortion.
The extrinsic parameters: the tranformation matrix between the 3D world and 3D camera coordinate systems.
The estimation of the above mentioned parameters is usually based on 2D-3D correspondences. The algorithm detects some 2D points in the image (i.e. chessboard) for what the corresponding 3D object points are specified (known 3D geometry). It performs the following steps in the simplest case (can vary on the flags of cv::calibrateCamera(..., int flags, ...)):
Computes the linear intrinsic parameters and considers the non-linear ones to zero.
Estimates the initial camera pose (extrinsics) in function of the approximated intrinsics. This is done using cv::solvePnP(...).
Performs the Levenberg-Marquardt optimization algorithm to minimize the re-projection error between the detected 2D image points and 2D projections of the 3D object points. This is done using cv::projectPoints(...).
cv::solvePnP(...)
At this point, I also answered implicitly the role of cv::solvePnP(...) as this is the part of cv::calibrateCamera(...).
Once you have the intrinsics of a camera, you can assume that these will never change (except you change the optics or zooming). On the other hand the extrinsics can be changed, i.e. you can rotate the camera or put it to another location. You should see that the scenario of changing an object's pose to the camera is very similar in this case. And this is what the cv::solvePnP(...) is used for.
The function estimates the object pose given:
A set of 3D object points in a model coordinate system (can be the 3D world as well),
Their 2D projections on the image plane,
The linear and non-linear intrinsic parameters.
The output of cv::solvePnP(...) is given as a rotation vector (rvec) together with a translation vector (tvec) that bring the 3D object points from the model coordinate system to the 3D camera coordinate system.
calibrateCamera (doc) estimates intrinsics coefficients (i.e. camera matrix and distortion coefficients) for a given camera. This function requires you to provide as input N sets of 2D-3D correspondences, associated to N images taken with the same camera from varying viewpoints (typically N=30, see this tutorial on this topic). The function returns the camera matrix and distortion coefficients for the considered camera. Although those are usually not used, the extrinsics parameters (i.e. position and orientation) are also estimated, hence the function returns one pair of rvec and tvec for each of the N input images.
solvePnP (doc) estimates extrinsics parameters for a given camera image. This function requires you to provide a set of 2D-3D correspondences, associated to a single image taken with a camera with known intrinsics parameters. The function returns a single pair of rvec and tvec, corresponding to the input image.
calibrateCamera() provides rvec, tvec, distCoeff, cameraMatrix ---- distCoeffs are related to distortion of the image and cameraMatrix provides the center of image(Cx and Cy) and focal length (Fx and Fy) (projection center). These are called intrinsic parameters. Unless you change the aperture/focus of the camera they will remain the same. [it also provides rvec and tvec, I don't know yet now what can be any possible use of it. These are the position of the camera in the real world. rvec and tvec are also known as extrinsic parameters]
solvePnP() takes cameraMatrix, distCoeff as input and provides rvec, tvec --- Using the Cx, Cy, Fx, Fy it can estimate the current position of the camera i.e. the extrinsic parameters.
In other words, first use calibrateCamera() to obtain the CameraMatrix and distCoeff. Use them in solvePNP() and it will tell you the rotation (rvec) and translation (tvec) of the camera as you move the camera with respect to your real world object (with some marker as you can presume).
Localization of an object specified in the image.
I am working on the project of computer vision to find the distance of an object using stereo images.I followed the following steps using OpenCV to achieve my objective
1. Calibration of camera
2. Surf matching to find fundamental matrix
3. Rotation and Translation vector using svd as method is described in Zisserman and Hartley book.
4. StereoRectify to get the projection matrix P1, P2 and Rotation matrices R1, R2. The Rotation matrices can also be find using Homography R=CameraMatrix.inv() H Camera Matrix.
Problems:
i triangulated point using least square triangulation method to find the real distance to the object. it returns value in the form of [ 0.79856 , .354541 .258] . How will i map it to real world coordinates to find the distance to an object.
http://www.morethantechnical.com/2012/01/04/simple-triangulation-with-opencv-from-harley-zisserman-w-code/
Alternative approach:
Find the disparity between the object in two images and find the depth using the given formula
Depth= ( focal length * baseline ) / disparity
for disparity we have to perform the rectification first and the points must be undistorted. My rectification images are black.
Please help me out.It is important
Here is the detail explanation of how i implemented the code.
Calibration of Camera using Circles grid to get the camera matrix and Distortion coefficient. The code is given on the Github (Andriod).
2.Take two pictures of a car. First from Left and other from Right. Take the sub-image and calculate the -fundmental matrix- essential matrix- Rotation matrix- Translation Matrix....
3.I have tried to projection in two ways.
Take the first image projection as identity matrix and make a second project 3x4d through rotation and translation matrix and perform Triangulation.
Get the Projection matrix P1 and P2 from Stereo Rectify to perform Triangulation.
My object is 65 meters away from the camera and i dont know how to calculate this true this based on the result of triangulation in the form of [ 0.79856 , .354541 .258]
Question: Do i have to do some extra calibration to get the result. My code is not based to know the detail of geometric size of the object.
So you already computed the triangulation? Well, then you have points in camera coordinates, i.e. in the coordinate frame centered on one of the cameras (the left or right one depending on how your code is written and the order in which you feed your images to it).
What more do you want? The vector length (square root of the sum of the square coordinates) of those points is their estimated distance from the same camera. If you want their position in some other "world" coordinate system, you need to give the coordinate transform between that system and the camera - presumably through a calibration procedure.
I am doing stereo calibration of two cameras (let's name them L and R) with opencv. I use 20 pairs of checkerboard images and compute the transformation of R with respect to L. What I want to do is use a new pair of images, compute the 2d checkerboard corners in image L, transform those points according to my calibration and draw the corresponding transformed points on image R with the hope that they will match the corners of the checkerboard in that image.
I tried the naive way of transforming the 2d points from [x,y] to [x,y,1], multiply by the 3x3 rotation matrix, add the rotation vector and then divide by z, but the result is wrong, so I guess it's not that simple (?)
Edit (to clarify some things):
The reason I want to do this is basically because I want to validate the stereo calibration on a new pair of images. So, I don't actually want to get a new 2d transformation between the two images, I want to check if the 3d transformation I have found is correct.
This is my setup:
I have the rotation and translation relating the two cameras (E), but I don't have rotations and translations of the object in relation to each camera (E_R, E_L).
Ideally what I would like to do:
Choose the 2d corners in image from camera L (in pixels e.g. [100,200] etc).
Do some kind of transformation on the 2d points based on matrix E that I have found.
Get the corresponding 2d points in image from camera R, draw them, and hopefully they match the actual corners!
The more I think about it though, the more I am convinced that this is wrong/can't be done.
What I am probably trying now:
Using the intrinsic parameters of the cameras (let's say I_R and I_L), solve 2 least squares systems to find E_R and E_L
Choose 2d corners in image from camera L.
Project those corners to their corresponding 3d points (3d_points_L).
Do: 3d_points_R = (E_L).inverse * E * E_R * 3d_points_L
Get the 2d_points_R from 3d_points_R and draw them.
I will update when I have something new
It is actually easy to do that but what you're making several mistakes. Remember after stereo calibration R and L relate the position and orientation of the second camera to the first camera in the first camera's 3D coordinate system. And also remember to find the 3D position of a point by a pair of cameras you need to triangulate the position. By setting the z component to 1 you're making two mistakes. First, most likely you have used the common OpenCV stereo calibration code and have given the distance between the corners of the checker board in cm. Hence, z=1 means 1 cm away from the center of camera, that's super close to the camera. Second, by setting the same z for all the points you are saying the checker board is perpendicular to the principal axis (aka optical axis, or principal ray), while most likely in your image that's not the case. So you're transforming some virtual 3D points first to the second camera's coordinate system and then projecting them onto the image plane.
If you want to transform just planar points then you can find the homography between the two cameras (OpenCV has the function) and use that.
So I have a depth map and the extrinsics and intrinsics of the camera.I want to get back the 3D points and the surface normals .I am using the functionReprojectImageTo3D.In the stereo rectify function to find Q how do I get the The rotation matrix
between
the 1st and the 2nd cameras’ coordinate systems? I have individual rotation matrix and translation vector but how do I get it for "between the cameras?"
.Also this would give me the 3D points .Is there a method to generate the surface normals?
Given that you have the extrinsic matrix of both cameras, can't you simply take the inverse extrinsic matrix of camera 1, multiplied by the extrinsic matrix of camera 2?
Also, for a direct relation between the two cameras, take a look at the Fundamental Matrix (or, more specific, the Essential matrix). See if you can find a copy of the book Multiple View Geometry by Hartley and Zisserman.
As for the surface normals, you can compute those yourself by computing crossproducts on the corners of triangles. However, you then first need the reconstructed 3D point cloud.