A C++17 wrapper for Nvidia Argus with support for zero-copy frame transfers to CUDA kernels and CUDA-accelerated AprilTag detection with ISAAC (no ROS required).
- Modern C++ wrapper for Argus that makes it easy to get frames to a CUDA kernel as fast as possible, with no copying--I read Nvidia's examples and the Argus headers so you don't have to!
- Example integration with the ISAAC GPU-accelerated AprilTag detector that does not require ROS
This is excerpted from a closed-source SLAM system based on UcoSLAM that I wrote (called "cucoslam" for CUDA and UcoSLAM; this is why the top-level namespace is cuco). I took all the commits up to a certain data that are relevant to someone trying to use Argus with CUDA or trying to use Argus with the ISAAC AprilTag detector. This isn't really ready to be a nice library that you can plug and play with, but I'd like to think that it's pretty close. It should at least serve as a good starting point.
For the moment this requires Argus, and it will always require a valid CUDA installation (11 or later). For these two reasons we currently only support building on the Jetson Nano's Jetpack 4.3 image. You'll also need to install libopencv-dev through apt. All other dependencies are packaged internally.
Note: the NVIDIA AprilTag library is currently required but not included for licensing reasons. The only version I've found that has the right device-side symbols for use indepedant of ISAAC is here. You'll have to place it in the project manually.