Hello All,
We have a use-case where we are using Fundamental Matrix API. We tried Fundamental Matrix Computation API on both x86 and TI, where we gave same set of input to both. The x86 version is the one available with OpenCV and the one on TI is with VLIB Library. We are observing the differences in output between OpenCV API and VLIB API. Following are more details on this :
I have also attached the test data which we used for comparison.
API's for comparison :
| TI (VLIB) | x86 (OpenCV) |
| VLIB_findFundamentalMat |
findFundamentalMat
|
Input Configuration (API Parameters) :
| Configuration Name | VLIB_findFundamentalMat (TI VLIB) Config value | findFundamentalMat (X86 OpenCV) Config value |
| numfeatures | 319 (data shared below) | 319 (data shared below) |
| distThreshold |
0.9
|
0.9 |
| degenThreshold |
0.176
|
NA |
| numItr4RANSAC |
0x7FFFFFFF
|
NA |
|
method
|
0x1F
|
FM_RANSAC |
Input Data : (Note :The data attached is for Single Run ) [Attached a CSV File containing all the matched Image Coordinates]
I have attached an excel sheet containing all the correspondence's (matched pixel coordinates of previous and current images) .
Image Configuration:
Width * Height : 1280 * 720
Principal Point X : 640
Principal Point Y : 360
Focal Length X : 2710.56
Focal Length Y : 2698.07
Differences In Output :
| TI Fundamental Mat O/P (3 * 3 Matrix) | x86 Fundamental Mat O/P (3 * 3 Matrix) |
| [ -0.000000367, -0.000000030, -0.000000525, -0.000007015, -0.000000573, 0.000000592, 0.001185923, 0.000098398, -0.010825714 ] |
[ 6.150850283990125e-05, -0.00109855653582384, -0.7636175924823848, 0.001306624128030765, 0.0003766128779860109, -1.048454432812976, 0.6343923978386575, 0.979978637884417, 1 ] |
Attachment :
feature_matching_coordinates_count_293.csv
I hope the above data is sufficient for the problem statement. Please let us know if we are missing anything else. Your support is highly appreciated.
Thanks and Best Regards,
Neel
