hey everyone, im working on visual odometry project using a nicla vision board, i already did a script to do the feature extraction using ORB ( by find_keypoints() ) but i want to do the mapping now and for that i wanted to devide the frame preview in OPENMV IDE into 2 , right one is frame at time T and left one is frame T+1 and make it look like like this
Hi, the Nicla doesn’t really have enough memory to do what you want… You’ll need to create two temporary frame buffers. You can use the API in the sensor module to allocate a second and third frame buffer.
After you do this you can copy the current image into one of the temporary buffers, then apply the other buffer on half the image. Then you swap which of the two extra buffers was the oldest and do it again for the next frame.
The only issue is that you’re going to be very limited on the frame size here. Like QQVGA.
1 Like
yess thank you it worked actually and only working for QVGA and did some keypoints extractions on the 2 frames, but when i tried to do the estimation of the transformation matrixes, i needed to have K which is the calibration matrix of the nicla vision, tried to search how to calibrate an uncalibrated cam but i found lots of exemples where they would use opencv functions, Do you think its possible to find the intrinsic parameters of its camera,
Thank you in advance,
Oussama
Hi, camera lens correction is something we’ve needed to provide for a while now. However, it is not part of our SDK though.
Numpy is available on the cameras though, you can do much of the matrix math here. However, unless you know exactly what you are doing it’s not going to be easy.
Anyway, I don’t have a solution for you about this. It’s not impossible, the hardware/system can do it. But, you pretty much need to implement from scratch.
We are focused on launching new systems right now however, so, camera calibration is not my focus currently.