HPEQuick

Quick example of Human Pose Estimation using OpenVINO and an OpenPose model

Requirements

• OpenCV
• OpenVINO 2023.3

How to Build

First, install the requirements. On Unixes, brew or apt can be used to install OpenVINO.

On MacOS, OpenCV is better installed from source. Follow the instructions on the OpenCV documentation to install it: https://docs.opencv.org/master/d7/d9f/tutorial_linux_install.html. OpenVINO can be installed using brew install openvino.

On Linux, apt install libopencv-dev openvino-2023.3.0 is fine.

Then, on both MacOS and Linux, build the project using CMake (by default, the project is built in Release configuration):

cmake -Bbuild
cmake --build build

On Windows, The project is tested on Visual Studio 2022.

Follow the instructions on the OpenVINO documentation to install the toolkit: https://docs.openvino.ai/2023.3/openvino_docs_install_guides_overview.html. Likewise, for OpenCV follow instructions on https://docs.opencv.org/4.x/d3/d52/tutorial_windows_install.html.

With OpenVINO extracted to C:\Program Files (x86)\Intel\openvino_2023.3 and OpenCV to C:\OpenCV, remember to add to the PATH environment variable the dll directories, typically: C:\OpenCV\build\x64\vc16\bin, C:\Program Files (x86)\Intel\openvino_2023.3\runtime\bin\intel64\Release, and C:\Program Files (x86)\Intel\openvino_2023.3\runtime\3rdparty\tbb\bin.

Then, create the following environment variables to the paths of the respective installation dirs (once and for good):

setx OpenCV_DIR C:\OpenCV
setx OpenVINO_DIR 'C:\Program Files (x86)\Intel\openvino_2023.3'

Finally, configure the project and build it as follows:

cmake -Bbuild 
cmake --build build --config Release

How to Run

The program takes a single argument, which is the path to the OpenPOSE model to use for inference. The model is automatically downloaded by CMake in the models directory. Tu run the program, use the following command:

build/quick human-pose-estimation-0001.xml

If you have more than one camera, you can specify the camera index as a second argument:

build/quick human-pose-estimation-0001.xml 1

On a laptop, index 0 (default) corresponds to the embedded webcam. On MacOS, index 1 corresponds to the continuity camera, i.e. the iPhone camera connected on WiFi.