segmentation

A ROS package that converts images into a semantic scene graph for other systems.

Goals

The final deliverable should be a reusable modular ROS package that takes video feed as input and outputs labeled segmented scene semantics.

The goal for this repository is not a robot using this package. Integration will occur in another repository.

Timeline

Part 1: Convert video into a semantic scene

1. [X] Install YOLO, OpenCV, and ROS Noetic
2. Implement YOLO on a test image
3. Modify YOLO to work with a live camera feed (Webcam into ROS)
4. Implement deduplication of objects between frames
5. Implement data structure to store data compatible with ROS messages

Part 2: Enhance model to generate scene graphs

1. Read about Relational Scene Graphs by Jenkins
2. Implement scene graphs
3. Integrate with previous ROS system

Setup

Add the project to a catkin workspace.

rosdep install --from-paths src --ignore-src -r -y
catkin build
cd src/segmentation/object_detection
catkin test --this

You can also run the tests via docker:

docker compose build
docker compose run --rm app catkin test

Usage

Run a container based on image and attach a shell

docker run -it segmentation-app /bin/bash