Running the Object Detection Node

This guide explains how to run the object detection node from the obj_detect package, which provides real-time object detection capabilities using the HailoRT API and YOLOv11n model.

Overview

The object detection node (object_detector) is designed to:

Subscribe to camera image data from the /camera/image_raw topic
Perform real-time object detection using a YOLOv11n model optimized for Hailo hardware
Publish detection results to the /detect topic with bounding boxes and class information
Detect two specific classes: Purple ball and Green ball

Prerequisites

Before running the object detection node, ensure you have:

ROS2 Environment: The workspace must be built and sourced
HailoRT API: Required for running the YOLOv11n model on Hailo hardware
Model File: The yolov11n_2cls.hef model file should be available
Camera Data: A camera node publishing to /camera/image_raw topic (See Camera Operations)[../camera/camera_operations.md]

Building the Package

Change into the common_platform_ws directory

cd ~/repos/common_platform/common_platform_ws

Only do this once - build the package

colcon build --packages-select obj_detect --symlink-install

Source the package

source install/setup.bash

To make things easier, after you have built the package, you can combine these into two commands to switch into this repository and source the setup.bash (this launches a virtual environment). Once you have sourced the environment, you can then launch the object detector.

cd ~/repos/common_platform/common_platform_ws
source install/setup.bash

Running the Object Detection Node

Method 1: Using Launch File (Recommended)

The easiest way to run the object detection node is using the provided launch file:

ros2 launch obj_detect object_detector.launch.py

This will start the node with default parameters:

Model path: /home/rcr/repos/common_platform/models/yolov11n_2cls.hef
Confidence threshold: 0.5
Input dimensions: 640x640
Log directory: /tmp

Method 2: Custom Parameters

You can override default parameters when launching:

ros2 launch obj_detect object_detector.launch.py \
  model_path:=/path/to/your/model.hef \
  confidence_threshold:=0.6 \
  input_width:=640 \
  input_height:=640 \
  hailort_log_path:=/path/to/logs

Method 3: Direct Node Execution

To run the node directly without the launch file:

ros2 run obj_detect object_detector

Note: When running directly, you’ll need to set parameters manually or use the launch file for proper configuration.

Configuration Parameters

Parameter	Default Value	Description
`model_path`	`/home/rcr/repos/common_platform/models/yolov11n_2cls.hef`	Path to the HEF model file
`confidence_threshold`	0.5	Minimum confidence for detections (0.0-1.0)
`input_width`	640	Input image width in pixels
`input_height`	640	Input image height in pixels
`hailort_log_path`	`/tmp`	Directory for HailoRT log files

Topics

Subscribed Topics

/camera/image_raw (sensor_msgs/Image): Input camera images for object detection

Published Topics

/detect (obj_detect/DetectionArray): Detection results containing:
- Bounding box coordinates (x, y, width, height)
- Class names (“Purple ball” or “Green ball”)
- Confidence scores

Message Format

The detection results are published using custom message types:

Detection.msg:

string class_name    # "Purple ball" or "Green ball"
float32 confidence   # Detection confidence (0.0-1.0)
float32 x           # Bounding box center x-coordinate
float32 y           # Bounding box center y-coordinate
float32 width       # Bounding box width
float32 height      # Bounding box height

DetectionArray.msg:

std_msgs/Header header
Detection[] detections

Verification and Monitoring

Check Node Status

ros2 node list
ros2 node info /rcr0nn/object_detector

Monitor Detection Results

# View detection messages
ros2 topic echo /rcr0nn/detect

# Check topic information
ros2 topic info /rcr0nn/detect
ros2 topic hz /rcr0nn/detect

Verify Camera Input

# Check if camera is publishing
ros2 topic list | grep camera
ros2 topic hz /rcr0nn/camera/image_raw

Troubleshooting

Common Issues

Model File Not Found
- Ensure the model file exists at the specified path
- Check file permissions
No Camera Data
- Verify camera node is running and publishing to /camera/image_raw
- Check camera topic with ros2 topic list
HailoRT Errors
- Check HailoRT installation and hardware connection
- Review log files in the specified hailort_log_path
No Detections
- Lower the confidence threshold
- Ensure objects are within the camera’s field of view
- Check lighting conditions

Logging

HailoRT logs are written to the specified directory (default: /tmp). The log file hailort.log contains detailed information about the inference process.

To use a custom log directory:

ros2 launch obj_detect object_detector.launch.py hailort_log_path:=/your/custom/directory

Integration with Other Nodes

The object detection node is designed to work seamlessly with other ROS2 nodes:

Camera nodes: Publishes to /camera/image_raw
Navigation nodes: Can subscribe to /detect for obstacle avoidance
Visualization nodes: Can display bounding boxes on camera feed
Control nodes: Can use detection data for robot control decisions

Performance Considerations

The node processes images at the camera’s publishing rate
Higher confidence thresholds reduce false positives but may miss valid detections
Input image dimensions affect processing speed and accuracy
HailoRT hardware acceleration provides real-time performance

This completes the setup for running the object detection node. The system is now ready to detect Purple and Green balls in real-time camera feeds.