View my account

Where do the post-processing codes included in pre-set run?

Comments

1 comment

  • MartyG

    The Vision Processor D4 circuit board inside the camera performs adjustments such as "rectification" of the captured raw camera data and then sends the updated data through the USB cable to the computing device.  The D4 hardware does processing that a graphics GPU in a computing device would normally do and so enables the camera to work with low-specification devices.

    Post-processing is applied after the data is initially adjusted by the D4 though and so will consume resources on the computing device.  Intel's post-processing documentation contains a section on computational burden, linked to here:

    https://dev.intelrealsense.com/docs/depth-post-processing#section-example-results-and-trade-offs 

    CPU workload can be reduced by "offloading" some of the work onto the GPU.  On devices with an Nvidia graphics GPU, you can build Librealsense with CUDA support.  For other GPU brands the SDK supports offloading with "GLSL Processing Blocks", though it may be ineffective on low-power computing devices.

    https://github.com/IntelRealSense/librealsense/issues/4905#issuecomment-533854888 

    Another approach to accelerating the processing of computer vision applications is to insert an Intel Neural Compute Stick 2 hardware acceleration device into a USB 3.0 port on the computing device alongside the RealSense camera.

    https://store.intelrealsense.com/buy-intel-neural-compute-stick-2.html 

    0
    Comment actions Permalink

Please sign in to leave a comment.