djkpA's profile - activity

How do I get R, t for projectPoints. Do I need to measure them manually? Is there any tool to simulate a 3D model and generate that data?

Hello , I have used solvePnP to find the Pose of object and I am getting some results for rvet and tvet. Now I wanted to know how accurate they are.

How to I compute the accuracy of retunred values of solvePnP? 1 method I found was re-projection error.

But is there any way to generate test cases:

Data need to generate as test cases:

(image points, object points)

(expected rvet, expected tvet)

Result: (computed rvet, computed tvet) - return values from solvePnP with each test case that we generated.

And now comparing the ((expected rvet, expected tvet) and (computed rvet, computed tvet) to measure the accuracy for different flags available for solvePnP.

Are there any ways/software/tools that helps me to generate accurate test cases.(Varieties of test cases may include nose environment, distance of object from camera varies, planar object points, non co-planar object points .. etc) ?

In solvePnP code link I mentioned in line 95 it is calling undistortPoints(ipoints, undistortedPoints, cameraMatrix, distCoeffs); I think the error is from this call

OpenCV Error:

Assertion failed (CV_IS_MAT(_src) && CV_IS_MAT(_dst) && (_src->rows == 1 || _src->cols == 1) && (_dst->rows == 1 || _dst->cols == 1) && _src->cols + _src->rows - 1 == _dst->rows + _dst->cols - 1 && (CV_MAT_TYPE(_src->type) == CV_32FC2 || CV_MAT_TYPE(_src->type) == CV_64FC2) && (CV_MAT_TYPE(_dst->type) == CV_32FC2 || CV_MAT_TYPE(_dst->type) == CV_64FC2)) in cvUndistortPoints, file /home/javvaji/opencv-3.2.0/modules/imgproc/src/undistort.cpp, line 293

retval, rvec, tvec = cv2.solvePnP(cam.object_points, cam.image_points, cam.camera_matrix, cam.dist_coefficients, None, None, False, cv2.SOLVEPNP_P3P)

Parameter values: (all are numpy arrays)

 Image points: 
[[ 433.   50.]
 [ 512.  109.]
 [ 425.  109.]
 [ 362.  106.]]
Object points: 
[[ 0.   0.   0. ]
 [ 6.5  0.   0. ]
 [ 0.   0.   6.5]
 [ 0.   6.5  0. ]]
Cam mat:
[[ 811.13275146    0.          322.47589111]
 [   0.          811.27490234  225.78684998]
 [   0.            0.            1.        ]]
Dist Coff:
[[-0.07649349 -0.02313312 -0.00911118 -0.0049251   0.74063563]]

I am using solvePnP function with flag SOLVEPNP_P3P. It is giving assertion error. The same solvePnP code works fine with SOLVEPNP_ITERATIVE flag. With P3P flag it internally calls undistortPoints function which is giving error.

solvePnP code ref: https://github.com/opencv/opencv/blob...

How to resolve this?

Hello I am trying to get the Pose of known geometry object and I am using opencv3.2.0 ,solvePnP to find that.

While searching in google I found that there are several algorithms for solvePnP implementation which we are passing as flag parameter to that function. But I did not find the requirements of each algorithm and when should I use which and I am trying to track the in real time.

Flags available for solvePnP:

SOLVEPNP_ITERATIVE

SOLVEPNP_P3P

SOLVEPNP_EPNP

SOLVEPNP_DLS (in opencv3)

SOLVEPNP_UPNP (in opencv 3)

What are the minimum no of points required for each algorithm?

Can those points be coplanar and noncoplanar ?

What is the time complexity and accuracy of each algorithm ?

I am getting 3d points of object in real word with triangulatePoints function in opencv. But to debug and verify accuracy of point I need a debugging procedure. I am gets new point for every 20 mill seconds. I need to plot the points and visualize them in real time.

How do I achieve this. I am developing in python.

Even if I pass np.float32([[x,y]]) as parameter to triangulatePointsit still gives same error @berak

xy is single point @berak

Currently it is xy = [x,y] and passing np.float32(xy) as parameter. But I changed it and checked with xy= np.float32([x,y]) and passing np.float32(xy) as parameter. Still error exists.

@berak while passing to triangulatePoints I am pasing as np.float32(xy1) and np.float32(xy2). and 3x4 matrix is filled by using solvePnP.

I am new to OpenCV. While using cv2.triangulatePoints I am getting the error mentioned below.

Error: OpenCV Error: Unsupported format or combination of formats (Input parameters must be matrices) in cvTriangulatePoints, file /io/opencv/modules/calib3d/src/triangulate.cpp, line 64

cv2.version '3.2.0'

triangulationOutput = cv2.triangulatePoints(projection_matrix_cam1, projection_matrix_cam2, np.float32(xy1), np.float32(xy2))

Declaration of params:

projection_matrix_cam1 = np.array([], dtype=np.float32).reshape(0,3,4) #lly projection_matrix_cam2
xy1 = [x,y] #lly xy2 , x,y are integers

Type of params respectively:

<type 'numpy.ndarray'>
<type 'numpy.ndarray'>
<type 'numpy.ndarray'>
<type 'numpy.ndarray'>

How to resolve this.

Hi everyone! I'm totally newbie at opencv and computer vision.I am trying to implement the SDK for Augmented reality.

Consider an example of Augmented reality book. I know the book that is used for that particular AR application.So now I can develop images from pages of the book(each page as each image and some parts of page as an individual image).Now using some OpenCV algorithms with the help of developed image data set I need to detect that image when viewing from camera in real time. I need to be able to tract the detected image until that is in my camera point of view viewed from any angle.

Could anyone please suggest some best algorithms/procedures that suits for my problem. Should I consider the Image data set size(no of images) in choosing right algorithm?