I'm trying to estimate the relative camera pose using OpenCV. Cameras in my case are calibrated (i know the intrinsic parameters of the camera).
Given the images captured at two positions, i need to find out the relative rotation and translation between two cameras. Typical translation is about 5 to 15 meters and yaw angle rotation between cameras range between 0 - 20 degrees.
For achieving this, following steps are adopted.
a. Finding point corresponding using SIFT/SURF
b. Fundamental Matrix Identification
c. Estimation of Essential Matrix by E = K'FK and modifying E for singularity constraint
d. Decomposition Essential Matrix to get the rotation, R = UWVt or R = UW'Vt (U and Vt are obtained SVD of E)
e. Obtaining the real rotation angles from rotation matrix
Experiment 1: Real Data
For real data experiment, I captured images by mounting a camera on a tripod. Images captured at Position 1, then moved to another aligned Position and changed yaw angles in steps of 5 degrees and captured images for Position 2.
Problems/Issues:
Sign of the estimated yaw angles are not matching with ground truth yaw angles. Sometimes 5 deg is estimated as 5deg, but 10 deg as -10 deg and again 15 deg as 15 deg.
In experiment only yaw angle is changed, however estimated Roll and Pitch angles are having nonzero values close to 180/-180 degrees.
Precision is very poor in some cases the error in estimated and ground truth angles are around 2-5 degrees.
How to find out the scale factor to get the translation in real world measurement units?
The behavior is same on simulated data also.
Have anybody experienced similar problems as me? Have any clue on how to resolve them.
Any help from anybody would be highly appreciated.
(I know there are already so many posts on similar problems, going trough all of them has not saved me. Hence posting one more time.)
In chapter 9.6 of Hartley and Zisserman, they point out that, for a particular essential matrix, if one camera is held in the canonical position/orientation, there are four possible solutions for the second camera matrix: [UWV' | u3], [UWV' | -u3], [UW'V' | u3], and [UW'V' | -u3].
The difference between the first and third (and second and fourth) solutions is that the orientation is rotated by 180 degrees about the line joining the two cameras, called a "twisted pair", which sounds like what you are describing.
The book says that in order to choose the correct combination of translation and orientation from the four options, you need to test a point in the scene and make sure that the point is in front of both cameras.
For problems 1 and 2,
Look for "Euler angles" in wikipedia or any good math site like Wolfram Mathworld. You would find out the different possibilities of Euler angles. I am sure you can figure out why you are getting sign changes in your results based on literature reading.
For problem 3,
It should mostly have to do with the accuracy of our individual camera calibration.
For problem 4,
Not sure. How about, measuring a point from camera using a tape and comparing it with the translation norm to get the scale factor.
Possible reasons for bad accuracy:
1) There is a difference between getting reasonable and precise accuracy in camera calibration. See this thread.
2) The accuracy with which you are moving the tripod. How are you ensuring that there is no rotation of tripod around an axis perpendicular to surface during change in position.
I did not get your simulation concept. But, I would suggest the below test.
Take images without moving the camera or object. Now if you calculate relative camera pose, rotation should be identity matrix and translation should be null vector. Due to numerical inaccuracies and noise, you might see rotation deviation in arc minutes.
Related
I'm new to opencv and computer vision. I want to find the R and t matrix between two camera pose. So I generally follows the wikipedia:
https://en.wikipedia.org/wiki/Essential_matrix#Determining_R_and_t_from_E
I find a group of the related pixel location of the same point in the two images. I get the essential matrix. Then I run the SVD. And print the 2 possible R and 2 possible t.
[What runs as expected]
If I only change the rotation alone (one of roll, pitch, yaw) or translation alone (one of x, y, z), it works perfect. For example, if I increase pitch to 15 degree, then I would get R and find the delta pitch is +14.9 degree. If I only increase x for 10 cm, then the t matrix is like [0.96, -0.2, -0.2].
[What goes wrong]
However, if I change both rotation and translation, the R and t is non-sense. For example, if I increase x for 10 cm and increase pitch to 15 degree, then the delta degree is like [-23, 8,0.5], and the t matrix is like [0.7, 0.5, 0.5].
[Question]
I'm wondering why I could not get a good result if I change the rotation and translation at the same time. And it is also confusing why the unrelated rotation or translation (roll, yaw, y, z) also changes so much.
Would anyone be willing to figure me out? Thanks.
[Solved and the reason]
OpenCV use a right-hand coordinate system. This is to say that the z-axis is projected from xy plane to the viewer direction. And our system is using a left-hand coordinate system. So as long as the changes are related to the z-axis, the result is non-sense.
This is solved due to the difference of the coordinates using.
OpenCV use a right-hand coordinate system. This is to say that the z-axis is projected from xy plane to the viewer direction. And our system is using a left-hand coordinate system. So as long as the changes are related to the z-axis, the result is non-sense.
I'm working on an object tracking application using openCV. I want to convert my pixel coordinates to world coordinates to get more meaningful information. I have read a lot about computing the perspective transform matrix, and I know about cv2.solvePnP. But I feel like my case should be special, because I'm tracking a runner on a track and field runway with the runway orthogonal to the camera's z-axis. I will set up the camera to ensure this.
If I just pick two points on the runway edge, I can calculate a linear conversion from pixels to world coords at that specific height (ground level) and distance from the camera (i.e. along that line). Then I reason that the runner will run on a line parallel to the runway at a different height and slightly different distance from the camera, but the lines should still be parallel in the image, because they will both be orthogonal to the camera z-axis. With all those constraints, I feel like I shouldn't need the normal number of points to track the runner on that particular axis. My gut says that 2-3 should be enough. Can anyone help me nail down the method here? Am I completely off track? With both height and distance from camera essentially fixed, shouldn't I be able to work with a much smaller set of correspondences?
Thanks, Bill
So, I think I've answered this one myself. It's true that only two correspondence points are needed given the following assumptions.
Assume:
World coordinates are set up with X-axis and Y-axis parallel to the ground plane. X-axis is parallel to the runway.
Camera is translated and possibly rotated about X-axis (angled downward), but no rotation around Y-axis(camera plane parallel to runway and x-axis) or Z-axis (camera is level with respect to ground).
Camera intrinsic parameters are known from camera calibration.
Method:
Pick two points in the ground plane with known coordinates in world and image. For example, two points on the runway edge as mentioned in original post. The line connecting the poitns in world coordinates should not be parallel with either X or Z axis.
Since Y=0 for these points, ignore the second column of the rotation/translation matrix, reducing the projection to a planar homography transform (3x3 matrix). Now we have 9 degrees of freedom.
The rotation assumptions will enforce a certain form on the rotation/translation matrix. Namely, the first column and first row will be the identity (1,0,0). This further reduces the number of degrees of freedom in the matrix to 5.
Constrain the values of the second column of the matrix such that cos^2(theta)+sin^2(theta) = 1. This reduces the number of unknowns to only 4. Two correspondence points will give us the 4 equations we need to calculate the homography matrix for the ground plane.
Factor out the camera intrinsic parameter matrix from the homography matrix, leaving the rotation/translation matrix for the ground plane.
Due to the rotation assumptions made earlier, the ignored column of the rotation/translation matrix can be easily constructed from the third column of the same matrix, which is the second column in the ground plane homography matrix.
Multiply back out with the camera intrinsic parameters to arrive at the final universal projection matrix (from only 2 correspondence points!)
My test implentation has worked quite well. Of course, it's sensitive to the accuracy of the two correspondence points provided, but that's kind of a given.
I am trying to use the PNP algorithm implementations in Open CV (EPNP, Iterative etc.) to get the metric pose estimates of cameras in a two camera pair (not a conventional stereo rig, the cameras are free to move independent of each other). My source of images currently is a robot simulator (Gazebo), where two cameras are simulated in a scene of objects. The images are almost ideal: i.e., zero distortion, no artifacts.
So to start off, this is my first pair of images.
I assume the right camera as "origin". In metric world coordinates, left camera is at (1,1,1) and right is at (-1,1,1) (2m baseline along X). Using feature matching, I construct the essential matrix and thereby the R and t of the left camera w.r.t. right. This is what I get.
R in euler angles: [-0.00462468, -0.0277675, 0.0017928]
t matrix: [-0.999999598978524; -0.0002907901840156801; -0.0008470441900959029]
Which is right, because the displacement is only along the X axis in the camera frame. For the second pair, the left camera is now at (1,1,2) (moved upwards by 1m).
Now the R and t of left w.r.t. right become:
R in euler angles: [0.0311084, -0.00627169, 0.00125991]
t matrix: [-0.894611301085138; -0.4468450866008623; -0.0002975759140359637]
Which again makes sense: there is no rotation; the displacement along Y axis is half of what the baseline (along X) is, so on, although this t doesn't give me the real metric estimates.
So in order to get metric estimates of pose in case 2, I constructed the 3D points using points from camera 1 and camera 2 in case 1 (taking the known baseline into account: which is 2m), and then ran the PNP algorithm with those 3D points and the image points from case 2. Strangely, both ITERATIVE and EPNP algorithms give me a similar and completely wrong result that looks like this:
Pose according to final PNP calculation is:
Rotation euler angles: [-9.68578, 15.922, -2.9001]
Metric translation in m: [-1.944911461358863; 0.11026997013253; 0.6083336931263812]
Am I missing something basic here? I thought this should be a relatively straightforward calculation for PNP given that there's no distortion etc. ANy comments or suggestions would be very helpful, thanks!
I’m working on an iOS augmented reality application.
It is location-based, not marker-based.
I use the GPS, compass and accelerometers to get latitude, longitude, altitude and the 3 euler angles: yaw, pitch and roll. I know using NSLog() that those 6 variables contain valid data.
My application shows some 3d objects over the camera view.
It works fine as long as I use everything but the roll angle.
If I add that third angle, the rotation applied to my opengl world is not good. I do it that way in the main OpenGL draw method
glRotatef(pitch, 1, 0, 0);
glRotatef(yaw, 0, 1, 0);
//glRotatef(roll, 0, 0, 1);
I think there is something wrong with this approach but am certainly not a specialist. Maybe I should create some sort of unique rotation matrix rather than 3 different ones?
Maybe that’s not possible easily? After all most desktop video games, FPS and the like, just let the user change the yaw and the pitch using the mouse, so only 2 angles, not 3. But unlike the mouse, which is a 2d device, a phone used for augmented reality can move in any angles.
But then again, all AR tutorials I have seen online couldn’t handle ‘roll’ properly. ‘Rolling’ your phone would either completely mess AR stuff up or do nothing at all, using some roll-compensation strategies.
So my question is, assuming I have my 3 Euler angles using the phone sensors, how should I apply them to my 3d opengl view?
I think you're likely talking about gimbal lock.
The essence of the problem is that if you rotate with Eulers then there's always a sequence to it. For example, you rotate around x, then around y, then z. But then one axis can always becomes ambiguous because a preceding can move it onto a different axis.
Suppose the rotation were 0 degrees around x, 90 degrees around y, then 20 degrees around z. So you do the x rotation and nothing has changed. You do the y rotation and everything moves 90 degrees. But now you've moved the z axis onto where the x axis was previously. So the z rotation will appear to be around x.
No matter what most people's instincts tell them, there's no way to avoid the problem. The kneejerk reaction is that you'll always rotate around the global axes rather than the local one. That doesn't resolve the problem, it just reverses the order. The z rotation could then the y rotation — which has already occurred — into an x rotation.
You're right that you should aim to create a unique description of rotation separated from measuring angles.
For augmented reality it's actually not all that difficult.
The accelerometer tells you which way down is. The compass tells you which way north is. The two may not be orthogonal though — the compass reading should vary from being exactly at a right angle to the floor on the equator to being exactly parallel to the accelerometer at the poles.
So:
just accept the accelerometer vector as down;
get the cross product of down and the compass vector to get your side vector — it should point along a line of longitude;
then get the cross product of your side vector and your down vector to get a north vector that is suitably perpendicular.
You could equally use the dot product to remove that portion of the compass vector that is in the direction of gravity and cross product from there.
You'll want to normalise everything.
That gives you three basis vectors, so just put them directly into a matrix. No further work required.
My problem statement is very simple. But I am unable to get the opencv calibration work for me. I am using the code from here : source code.
I have to take images parallel to the camera at a fixed distance. I tried taking test images (about 20 of them) only parallel to the camera as well as at different planes. Also I changed the size and the no of squares.
What would be the best way to calibrate in this scenario?
The undistorted image is cropped later, that's why it looks smaller.
After going through the images closely, the pincushion distortion seems to have been corrected. But the "trapezoidal" distortion still remains. Since the camera is mounted in a closed box, the planes at which I can take images is limited.
To simplify what Vlad already said: It is theoretically impossible to calibrate your camera with test images only parallel to the camera. You have to change your calibration board's orientation. In fact, you should have different orientation in each test image.
Check out the first two images in the link below to see how the calibration board should be slanted (or tilted):
http://www.vision.caltech.edu/bouguetj/calib_doc/
think about calibration problem as finding a projection matrix P:
image_points = P * 3d_points, where P = intrinsic * extrinsic
Now just bear with me:
You basically are interested in intrinsic part but the calibration algorithm has to find both intrinsic and extrinsic. Now, each column of projection matrix can be obtained if you select a 3D point at infinity, for example xInf = [1, 0, 0, 0]. This point is at infinity because when you transform it from homogeneous coordinates to Cartesian you get
[1/0, 0, 0]. If you multiply a projection matrix with a point at infinity you will get its corresponding column (1st for Xinf, 2nd for yInf, 3rd for zInf and 4th for camera center).
Thus the conclusion is simple - to get a projection matrix (that is a successful calibration) you have to clearly see points at infinity or vanishing points from the converging extensions of lines in your chessboard rig (aka end of the railroad tracks at the horizon). Your images don’t make it easy to detect vanishing points since you don’t slant your chessboard, nor rotate nor scale it by stepping back. Thus your calibration will always fail.