ArgusCameraOp

ArgusCameraOp captures a directly connected Jetson CSI camera through libArgus and an EGLStream. Argus supplies NV12 planes as CUDA arrays; the operator reads those arrays directly in CUDA and emits a display-ready BGRA tensor without an intermediate NV12 allocation or camera-to-CUDA copy.

Camera integration

For help connecting your Jetson CSI camera to CUDA processing and display, contact RidgeRun.

Input arguments

Argument	Meaning
`camera_index`	Zero-based Argus camera index.
`sensor_mode_index`	Sensor mode selected from the camera's Argus mode list.
`width`, `height`	Positive output dimensions requested from the selected mode.
`fps`	Positive capture rate used to configure the operator clock.
`frame_limit`	Optional limit used by the RidgeRun applications and validation runs.

Ports and data contract

Port	Memory	Queue	Trigger	Contract
`frame`	CUDA device, contiguous	N/A	`OnClock` at `fps`	Runtime-pool U8 tensor with shape `[height, width, 4]` in BGRA order. Frame identity and flow metadata are forwarded.

Build

Follow Holoscan RidgeRun Build and Installation for the module build and installation. The operator-specific include, target, and runtime requirements are listed below.

Include argus_camera/argus_camera.hpp and link the holoscan::argus_camera target. The operator requires Jetson Multimedia API Argus, nvargus_socketclient, CUDA and EGL on Jetson aarch64.

Example

The application must bind the output Tensor to the CUDA device used by the graph. A minimal display graph follows the repository applications:

auto camera = graph.op<ArgusCameraOp>("camera", 0U, 0U, 1920, 1080, 60);
auto display = graph.op<X11DisplayOp>("display", 1920, 1080, "Argus camera");
graph.add_flow(camera->frame, display->input);

compile_options.deployment.bind_tensor_output_device(
    holoscan::TensorOutputDevicePlacement{
        .operator_path = "camera",
        .output_port = "frame",
        .device = holoscan::DeviceId{0}});

The camera index and sensor-mode index are zero based. The selected sensor mode must report exactly the requested resolution, and the requested frame rate must fall within that mode's Argus frame-duration range. Width and height must be even because the EGLStream output is two-plane NV12.

Measurement scope

Argus includes Jetson ISP camera processing and produces NV12 before the CUDA conversion. Its GPU and latency measurements must not be compared as if they represented the same workload as a raw V4L2 Bayer pipeline.

Source reference

You can find the source code at ArgusCameraOp.

Related pages

V4L2UserPtrSourceOp documents direct RG10 capture.
X11DisplayOp and DrmBgraDisplayOp are compatible display endpoints.

❯