barracuda-vision

ROS 2 (Humble) YOLO object-detection node. It subscribes to an image topic, runs a local Ultralytics YOLO model on each frame, and publishes:

Topic	Type	Description
object_detector	std_msgs/Int32	Number of detections in the frame
detections	vision_msgs/Detection2DArray	Bounding boxes, class, confidence
detection_image	sensor_msgs/Image	Input frame annotated with boxes

It subscribes to yolo_input_image (sensor_msgs/Image).

Running with Docker

Step 1: Prepare the Model

Place your YOLO model file (e.g., yolov8n.pt) in a directory on your host, such as /localmodels. Only one model file should be present in this directory (the node refuses to start otherwise).

Step 2: Build and Run the Docker Container

The provided Dockerfile and docker-compose.yaml build the workspace with colcon and launch the node. Mount your model directory with the MODEL_PATH environment variable (defaults to /localmodels); it is mounted to /models inside the container, where the node looks for the model by default.

MODEL_PATH=/localmodels docker compose up --build

Step 3: Publish a Test Image

In another shell, exec into the container and run the test publisher, which publishes a static image to yolo_input_image:

docker exec -it barracuda-vision bash
source /opt/barracuda-vision/ros2_ws/install/setup.bash
ros2 run barracuda_vision test_pub_image

Step 4: View the Results

ros2 topic echo /detections          # bounding boxes
ros2 topic echo /object_detector     # detection count
rqt                                  # view /detection_image (Image View plugin)

Building Locally (without Docker)

Requires ROS 2 Humble plus ros-humble-cv-bridge and ros-humble-vision-msgs, and the Python deps in requirements.txt.

cd ros2_ws
colcon build --symlink-install
source install/setup.bash

# Launch the detector (override model location as needed)
ros2 launch barracuda_vision barracuda_vision.launch.py model_dir:=/path/to/models
# or point directly at a model file:
ros2 launch barracuda_vision barracuda_vision.launch.py model_path:=/path/to/model.pt

Parameters

yolo_image_listener:

• model_path (string, default ""): explicit path to a .pt model file.
• model_dir (string, default /models): directory scanned for exactly one model file when model_path is not set.
• input_topic (string, default yolo_input_image): image topic to subscribe to.

test_pub_image:

• image_path (string): image to publish (defaults to the bundled shark.png).
• topic (string, default yolo_input_image).
• rate (double, default 10.0): publish rate in Hz.

---

Note:

• Only local model inference is supported (no Roboflow / remote server).
• The node exits if zero or more than one model file is found in model_dir (when model_path is unset).