ROS YOLOv8 Object Detection Package

This ROS package utilizes YOLOv8 for real-time object detection in images received from a camera topic. The package is designed to detect specific objects, such as stop signs, tires, and pedestrians, based on user requests.

NOTE: This package is a prototype meant for use with the LTU ACTor Development Vehicle only. Using it on other systems may require significant modifications. Fully developed package is kept private to the LTU ACTor Organisation.

Prerequisites

• Python >= 3.8

• ROS Noetic (Robot Operating System)

• OpenCV

• Ultralytics YOLOv8

• PyTorch

• Numpy

• CUDA Toolkit >= 11(refer pytorch installation for best version) (not required if not using GPU)

• Yolov8 Models - (not uploaded here since they are big, you can train your own models and reference them from the /models folder)

Installation

1. Clone this repository to your ROS workspace's /src folder:

   git clone https://github.com/Aeolus96/yolov8_ros.git

2. Build the ROS package:

   cd <ros_workspace>
   catkin_make

Usage

1. Launch the ROS node:

   roslaunch ltu_actor_route_yolov8_detector yolov8_detector.launch

2. Subscribe to the relevant topics to trigger object detection, set these parameters during launch:

   • Image Topic: <imgtopic>
   • Object Detection Topic: <look_for_object_topic_name> (string type, add unique strings and models in detect.py as needed)

3. View the results:

   The package provides detection results for stop signs, tires, and pedestrians. The detected objects, along with their sizes, are published to specific topics.

4. Debugging:

   If debugging is enabled (set in the rqt dynamic reconfigure), annotated images showing the detected objects are published to debug topics.

Configuration

The package supports dynamic reconfiguration through the ROS Parameter Server. Parameters such as image resizing, flipping, and debugging can be adjusted on-the-fly.

Yolo v8 models can be added or swapped out in the /models folder and names updated in the launch file.

Topics

• Inputs:

  • Image Topic: <imgtopic_name>
  • Input String Topic: <look_for_object_topic_name>

• Stop Sign Detection:

  • Detected Topic: <stop_sign_detected_topic_name>
  • Size Topic: <stop_sign_size_topic_name>
  • Debug Topic: <stop_sign_debug_topic_name>

• Person Detection:

  • Detected Topic: <person_detected_topic_name>
  • Size Topic: <person_size_topic_name>
  • Debug Topic: <person_debug_topic_name>

• Tire Detection:

  • Detected Topic: <tire_detected_topic_name>
  • Size Topic: <tire_size_topic_name>
  • Debug Topic: <tire_debug_topic_name>

Contributing

Contributions are not allowed here. Fully developed and updated package is kept private to the LTU ACTor Organisation and only members are allowed to contribute.

License

This project is licensed under the MIT License.