Trolley Detection And Pose Estimation

A ROS package for real-time trolley detection and pose estimation using a YOLO ONNX model and RGB-D camera data.

The package performs:

• YOLO-based trolley and leg detection
• Depth-based validation of detections
• 3D localization of trolley legs
• Trolley pose estimation in the base_link frame
• Publishing of bounding boxes and goal pose for downstream navigation

Requirements

• ROS Noetic
• OpenCV
• ONNX Runtime (CPU)
• RGB-D camera (e.g., Intel RealSense)

---

CPU-Based Detection Setup

Step 1: Download and Install ONNX Runtime

VER=1.18.0
cd /tmp
wget https://github.com/microsoft/onnxruntime/releases/download/v$VER/onnxruntime-linux-x64-$VER.tgz
# Alternative: Get from misc folder (../../../../misc/libs/onnxruntime)

tar xf onnxruntime-linux-x64-$VER.tgz
sudo mv onnxruntime-linux-x64-$VER /opt/onnxruntime

Step 2: Configure Environment Variables

Add the following to your ~/.bashrc:

export ONNXRUNTIME_ROOT=/opt/onnxruntime
export LD_LIBRARY_PATH=$ONNXRUNTIME_ROOT/lib:$LD_LIBRARY_PATH

Then reload your shell:

source ~/.bashrc

Step 3: Build the Package

echo "/opt/onnxruntime/lib" | sudo tee /etc/ld.so.conf.d/onnxruntime.conf
sudo ldconfig

catkin config --init --install
catkin build trolley_detection

Step 4: Run

source /home/nuc/amr_ws/install/setup.bash
roslaunch trolley_detection trolley_detection.launch

---

Outputs

The node publishes:

• /trolley_box
• /trolley_all_boxes
• /trolley_base_pose_stamped
• /trolley_base_pose_stamped_original
• /detection_image_pre
• /detection_image_post

---

Author

Subhanshu Sethi