Hi,
That’s because the Jetson container doesn’t really include CUDA libraries but mount it from the host at the runtime.
Since the mounting is runtime-only, there is no CUDA library available at docker build.
Below are two possible workarounds for this issue:
1. Copy the CUDA toolkit into the container and delete it once the OpenCV is built.
2. You can also add a csv file to mount the host OpenCV library from Jetson directly.
/etc/nvidia-container-runtime/host-files-for-container.d
More, our user has successfully created an OpenCV Dockerfile based on l4t-base and sharing on the forum.
You can check his post to get some idea first:
Thanks.