OpenCV Accelerator

The RZ/V2H MPU platform includes the OpenCV Accelerator (OpenCVA), which uses the DRP IP to accelerate specific OpenCV functions and improve performance in computer vision applications.

Overview

The OpenCVA library is designed to automatically use the DRP to accelerate supported OpenCV functions when the conditions for DRP usage are met.

Note

The OpenCVA library version for the RZ/V2H RDK is 4.6.0, which is compatible with OpenCV version 4.6.0 on Ubuntu 24.04 LTS.

See also

For more detailed information about OpenCVA, refer to the RZ/V2H OpenCV Accelerator.

How to Install the OpenCVA Library

To install the OpenCVA library on the RZ/V2H RDK, follow these steps:

1. Remove any existing OpenCV library installed.

   If you have installed OpenCV using the apt package manager, you need to remove it before installing the OpenCVA library, because the OpenCVA library provides its own version of OpenCV that is optimized for the RZ/V2H RDK on Ubuntu 24.04.

   sudo apt remove -y libopencv* opencv* python3-opencv
   

2. Download and run the installation script:

   wget -qO install_opencv_arm64.sh https://raw.githubusercontent.com/renesas-rdk/rzv2h_opencv_accelerated_debs/main/install_opencv_arm64.sh
   sudo bash install_opencv_arm64.sh
   

   The script will install all OpenCVA Debian packages, resolve any missing dependencies, and verify the installation automatically.

   Note

   The installation process may report error about missing dependencies during the dpkg step. This is expected and will be resolved automatically by the script.

   After script execution, the OpenCVA library and its dependencies will be installed on your RZ/V2H RDK, and you can start using it in your computer vision applications.

How to Use OpenCVA

OpenCVA leverages the DRP’s processing capability to enhance specific functions of the OpenCV library.

When you call an OpenCV function that is supported by OpenCVA, the library automatically determines whether the DRP can be used to accelerate that function based on the input data and the current system conditions.

If the conditions under which DRP can be used are met, OpenCVA automatically executes the function using the DRP.

For the DRP usage conditions, see OpenCVA API Specification and Condition for Using DRP.

OpenCVA can also disable DRP on a per-function basis. See API Functions to Control OpenCVA for details.

The following table lists the OpenCV functions that can be executed using DRP in the OpenCVA:

OpenCV Functions Supported by DRP

OpenCV Function Name	Function
resize	Image resize.
cvtColor	Change color space.
cvtColorTwoPlane	Change color space.
GaussianBlur	Gaussian filter process.
dilate	Areas of bright regions grow.
erode	Areas of dark regions grow.
morphologyEX	Combination of dilate and erode.
filter2D	Image convolution.
Sobel	Extracting image edges.
adaptiveThreshold	Transforms a grayscale image to a binary image according to the formula.
matchTemplate	Compares a template against overlapped image regions.
wrapAffine	Transforms the source image using the 2x3 matrix.
wrapPerspective	Transforms the source image using the 3x3 matrix.
pyrDown	Downsampling step of the Gaussian pyramid construction.
pyrUp	Upsampling step of the Gaussian pyramid construction.
FAST	Detects corners using the FAST algorithm.
remap	Applies a generic geometrical transformation to an image.
StereoSGBM	Creates StereoSGBM object for the modified H. Hirschmuller algorithm.