pose-estimation is responsible for estimating the pose given a region with an operator. It uses the open-source pose estimator Openpose as its core engine.

This module has three core operations:

• Estimate
• Remove outsider operators
• Get pose position

Uses only one frame as input and returns the coordinates of every person in it, following a matrix of coordinates and probabilities, using the following schema:

It returns two things:

A 21 * 3 matrix with each pose coordinate and accuracy for each person detected

[[x, y, acc],
[x, y, acc], 
... ...
[x, y, acc],
[x, y, acc]]

• Ubuntu >= 18.04
• Nvidia Driver >= 410.68
• Docker & Nvidia-docker >= 20.10
• docker-compose >= 1.29.2
• RabbitMQ >= 3.11.17
• Kombu >= 5.1.0

Install nvidia-docker

sudo su
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu22.04/nvidia-docker.list > /etc/apt/sources.list.d/nvidia-docker.list
apt update
apt -y install nvidia-container-toolkit nvidia-docker2

• First, install Docker.

• Clone repository:

   git clone https://github.com/fmorenovr/openpose.git
  

Docker installation

We use docker in order to test locally, goes to the repository dir openpose. Then, first we need to run:

sudo docker-compose -f docker-compose.yaml up --build -d

Test the environment:

sudo docker exec -it openpose bash

Local Installation (Ubuntu 18.04)

• First, Install main libs:

# cmake, usually already installed on Linux
# cmake, usually already installed on Linux, in other case:
sudo apt-get install cmake cmake-qt-gui
sudo apt-get install build-essential libssl-dev
# other libraries
sudo apt-get install libgoogle-glog-dev
sudo apt-get install protobuf-compiler libprotobuf-dev
sudo apt-get install libboost-system-dev libboost-thread-dev libboost-program-options-dev libboost-test-dev libboost-filesystem1.74-dev
sudo apt-get install libatlas-base-dev
sudo apt-get install libhdf5-dev
sudo apt-get install libcaffe-cuda-dev
sudo apt-get install libgtk2.0-dev

• Second, install OpenPose following the installation instructions from here.

Step 1, Clone the OpenPose repository:

git clone https://github.com/CMU-Perceptual-Computing-Lab/openpose.git
cd openpose/
git submodule update --init --recursive --remote

Step-2, Config with CMAKE-GUI (local system only, not inside the docker):

mkdir build/
cd build/
cmake-gui ..

Step-2, Config CMAKE (terminal):

mkdir build/
cd build/
# CPU only
cmake -DBUILD_PYTHON=ON  -DGPU_MODE=CPU_ONLY .. && make -j `nproc`
# GPU 
cmake -DBUILD_PYTHON=ON  -DUSE_CuDNN .. && make -j `nproc`

Notes:

• You must ensure that BUILD_PYTHON is flagged on

• You can get the prototxt model files here. or just download models from:

  cd /app/openpose/models
  bash getModels.sh
  

Once you successfully installed OpenPose, copy the openpose/ directory inside this directory one.

Download test videos:

bash get_data.sh
python3 -m unittest discover -s ./openpose_test -p test_*.py

• Openpose repository https://github.com/CMU-Perceptual-Computing-Lab/openpose