This is a ROS2 package designed for generating segmentation masks from pixels having a specific color in images. It's useful for identifying and segmenting areas with uniform color in visual data: its primary design purpose is to generate masks from the gazebo ground plane, which typically has a uniform color.
Install the dependencies by running:
sudo apt-get install libopencv-dev ros-$ROS_DISTRO-cv-bridge ros-$ROS_DISTRO-vision-msgs
The Mono Color Segmentation node subscribers to an image topic, processing each received image, and then publishing a binary mask as a sensor_msgs/msg/Image with mono8 encoding and label information as a vision_msgs/msg/LabelInfo. It exposes the target_color parameter which takes in a vector of 3 uint8 specifying the RGB color that is to be segmented. The mask is generated using OpenCV's cv::inRange function, which sets to 255 all pixels that have the same color as target_color and to 0 all other pixels; the LabelInfo message thus contains the following class map:
- class_id: 0
class_name: others
- class_id: 255
class_name: ground_plane
This node uses generic topic names for its subscribers and publishers, however users are encouraged to use namespaces or remappings to integrate it with their own system. The package provides a dummy launch with the /intel_realsense_r200_depth namespace to be used with Nav2's turtlebot simulation:
ros2 launch mono_color_segmentation turtlebot_world_segmentation.launch.py/image_raw(sensor_msgs/msg/Image): Input image topic, should be a 3 channel color image.
/mask` (sensor_msgs/msg/Image): Binary mask with 255 (white) for pixels matching the target color and 0 (black) for all other pixels./label_info(vision_msgs/msg/LabelInfo): Class map that assigns human readable names to class ids.