Box filter changing size and image types

Autonomous Machines Jetson & Embedded Systems Jetson Xavier NX
#1

Hi,

Do you have any suggestions on the best way to program calculations on an image set? I want to create a continuous buffer of 5 images (FIFO) and then do pixel-wise calculations continuously on this set of 5 Frames to visualize them, is it possible to do this with available visionworks functions or would it require a custom approach? Also, for custom convolution/box filter only U8 image types are accepted as inputs, can a U16 be used for any of these or do you have any alternate suggestions to use this for a U16 image and not U8?

Thanking you in advance for your reply

#3

Hi,

Sorry for the late update.

We have another computer vision library called VPI, and it also supports the box filter.
Since VisionWorks is deprecated, you can give VPI a try:

May I know do you want to apply the filter cross-frame or inter-frame.
We don’t implement a cross-frame filter except for our DL inference engine(TensorRT).
But you can develop your own via the VPI host function listed below:

Thanks.

#4

Thank you AastaLLL, I want to calculate cross frame (5-10) statistics on each pixel, mean, standard deviation etc. I just wanted to figure out which approach would provide the best performance especially when calculating min,max,mean and standard deviation cross-frames.