I’m getting started and have spent some time looking at jetson-inference.
I have a camera that’s in motion and doesn’t always maintain the same orientation (by design). I know the angle in real time and would like to rotate the frames back to “zero” before running detect.
What’s the best way to do this?
In case anyone is in the same boat, I was able to get what I want with cudaWarpAffine from Jetson utils, and the Open CV docs for how to use warp affine for rotation. I can provide a code example if interested
ajkeeton, I have a similar challenge. I need to rotate a displayed video in real-time based on incoming accelerometer data to maintain an “upright” appearance. Did you find significant delay using cudaWarpAffine? Would you mind posting your code? Thanks!