Pedestrian Skeleton Detector

Maintained by NOVA

Overview

A ROS Node that uses YOLOv8 Pose Detection to detect pedestrian skeletons in real-time and publish the annotated results.

In:

  • /cameras/camera0 Image
    • Receives an image from the camera (topic: cameras/camera0).

Out:

  • /processed_image Image

    • Publishes the processed image with pose annotations (topic: processed_image).

  • /detection_status String

    • Publishes the detection status, including confidence in detecting pedestrians.

Function Descriptions

image_callback(self, msg: Image)

Processes an incoming camera image, applies YOLOv8 Pose Detection, and republishes the annotated image.

Steps:

  1. Converts the incoming ROS Image message into an OpenCV image using CvBridge.
  2. Runs YOLOv8 Pose Detection on the image.
  3. Extracts the detected pedestrian(s) and checks if confidence is above the threshold.
  4. If a pedestrian is detected, it publishes a message to the /detection_status topic.
  5. Annotates the image with pose keypoints.
  6. Converts the annotated image back into a ROS Image message.
  7. Publishes the annotated image to the /processed_image topic.
  8. Optionally, displays the processed image in a window for debugging.

Error Handling:

  • If an unsupported image encoding is detected, logs an error message.
  • If an exception occurs during processing, logs the error details.

Dependencies

  • rclpy: ROS 2 Python API
  • sensor_msgs.msg.Image: ROS Image message type
  • std_msgs.msg.String: ROS String message type for detection status
  • cv_bridge: Converts between ROS Image messages and OpenCV images
  • cv2: OpenCV for image processing
  • ultralytics.YOLO: YOLOv8 model for pose detection
  • numpy: Array operations

Usage

Launch the node and subscribe to /processed_image to receive annotated images and /detection_status for detection updates:

ros2 run <package_name> <node_executable>