TPU-Posenet on ROS2

Edge TPU Accelerator + Posenet + Python + Sync + LaptopPC "+ ROS2".
Inspired by google-coral/project-posenet and PINTO0309/TPU-Posenet.
This repository makes it possible to perform Posenet estimation using TPU on ROS2. Image messange publishing is slow (about 5Hz) because cv_bridge.cv2_to_imgmsg is slow on ROS2(ros2/rosidl_python#79). If you use opencv window, you can show fast results (about 50Hz) using single TPU.

0. Table of contents

1. Environment
2. Inference result
3. Introduction procedure
4. Usage
5. Reference articles

1. Environment

• Ubuntu 16.04
• (ROS 1 Kinetic) -> not need. But if you have already installed, you can ros1-ros2 bridge.
• ROS 2 Ardent
• OpenCV4.1.1-openvino
• Cv_bridge on python3
• Coral Edge TPU Accelerator
• USB Camera or ROSBAG
• Self-powered USB 3.0 Hub
• Python 3.5.2+

2. Inference result on RViz2

3. Introduction procedure

Install Edge TPU API and download models.

$ sudo apt-get update;sudo apt-get upgrade -y

$ sudo apt-get install -y python3-pip
$ sudo pip3 install pip --upgrade
$ sudo pip3 install numpy

$ wget https://dl.google.com/coral/edgetpu_api/edgetpu_api_latest.tar.gz -O edgetpu_api.tar.gz --trust-server-names
$ tar xzf edgetpu_api.tar.gz
$ sudo edgetpu_api/install.sh

$ git clone https://github.com/shunchan0677/TPU-Posenet-ROS2.git
$ cd TPU-Posenet-ROS2
$ cd models;./download.sh;cd ..

Install ROS2 and Cv_bridge.

### Install ROS2

sudo apt update && sudo apt install curl gnupg2 lsb-release
curl -s https://raw.githubusercontent.com/ros/rosdistro/master/ros.asc | sudo apt-key add -
sudo sh -c 'echo "deb [arch=amd64,arm64] http://packages.ros.org/ros2/ubuntu `lsb_release -cs` main" > /etc/apt/sources.list.d/ros2-latest.list'
export CHOOSE_ROS_DISTRO=ardent
sudo apt update

sudo apt install ros-$CHOOSE_ROS_DISTRO-desktop
sudo apt update
sudo apt install ros-$CHOOSE_ROS_DISTRO-rmw-opensplice-cpp # for OpenSplice
sudo apt install ros-$CHOOSE_ROS_DISTRO-rmw-connext-cpp # for RTI Connext (requires license agreement)

### for ros1-ros2 bridge
sudo apt update
sudo apt install ros-$CHOOSE_ROS_DISTRO-ros1-bridge


### Build cv_bridge for python3
git clone https://github.com/ros-perception/vision_opencv.git
cd vision_opencv
git checkout ros2
vim CMakeLists.txt # Change Boost REQUIRED python -> python3(line18, 22)

colcon build --symlink-install

4. Usage

Publish image topic on ROS2 (need usb-camera)

source /opt/ros/ardent/setup.bash
ros2 run image_tools cam2image

Subscrive image topic and Estimate pose

source /opt/ros/ardent/setup.bash
source /home/<user>/vision_opencv/install/local_setup.bash
python3 pose_camera_sigle_tpu.py

Show Result Image on Rviz2

source /opt/ros/ardent/setup.bash
rviz2

If you want to ROS2-ROS1 bridge...(need to launch roscore)

source /opt/ros/ardent/setup.bash
ros2 run ros1_bridge dynamic_bridge --bridge-all-topics

5. Reference articles

Edge TPU USB Accelerator analysis - I/O data transfer - Qiita - iwatake2222

[150 FPS ++] Connect three Coral Edge TPU accelerators and infer in parallel processing to get ultra-fast object detection inference performance ーTo the extreme of useless high performanceー - Qiita - PINTO

ROS index : Installing ROS2 via Debian Packages (https://index.ros.org/doc/ros2/Installation/Crystal/Linux-Install-Debians/#install-additional-packages-using-ros-1-packages)

ROS index : cv_bridge (https://index.ros.org//p/cv_bridge/github-ros-perception-vision_opencv/)

ROS 1とROS 2のつながり（ros1_bridge） (https://gbiggs.github.io/rosjp_ros2_intro/ros1_bridge.html)

image callback in python very slow(ros2/ros2#509)