edgeimpulse_ros

ROS2 Humble wrapper for Edge Impulse on Linux.

1. Topics

• /detection/input/image, image topic to analyze
• /detection/output/image, image with bounding boxes
• /detection/output/info, VisionInfo message
• /detection/output/results, results as text

2. Parameters

• frame_id (string), "base_link", frame id of output topics
• model.filepath (string), "", absolute filepath to .eim file
• show.overlay (bool), true, show bounding boxes on output image
• show.center (bool), false, show centroids on output image
• show.labels (bool), true, show labels on bounding boxes,
• show.classification_info (bool), true, show the attendibility (0-1) of the prediction

3. How to install

1. install edge_impulse_linux:
   pip3 install edge_impulse_linux

2. on some boards and aarch64 these are required (e.g. vm on mac m1):
   sudo apt-get install libatlas-base-dev libportaudio2 libportaudiocpp0 portaudio19-dev
pip3 install pyaudio
   

3. download your .eim file as "linux board" and choose your architecture

4. make your eim file as executable:
   cd /path/to/your/eim/file
chmod +x <file>.eim
   

5. clone this repo in your workspace:
   cd ~/dev_ws/src git clone https://github.com/gbr1/edgeimpulse_ros

6. check dependencies:
   cd ~/dev_ws
rosdep install --from-paths src --ignore-src -r -y
   

7. build:
   colcon build --symlink-install
source install/setup.bash
   

4. How to run

Launch the node:
ros2 run edgeimpulse_ros image_classification --ros-args -p model.filepath:="</absolute/path/to/your/eim/file.eim>" -r /detection/input/image:="/your_image_topic" `

5. Models

Here you find some prebuilt models: https://github.com/gbr1/edgeimpulse_example_models

6. Known issues

• if you use a classification model, topic results is empty
• you cannot change color of bounding boxes (coming soon)
• other types (imu and sound based ml) are unavailable

Copyright © 2022 Giovanni di Dio Bruno - gbr1.github.io