ROS wrapper for TensorRT 3D object detection using ZED SDK
To use:
clone into src folder
git clone https://github.com/terrytao19/zed_ros_3d.git
├── ros_ws
│ ├── src
│ │ ├── package_1
│ │ ├── package_2
│ │ ├── zed_ros_3d
Make sure you have zed_sdk installed
Build package
catkin build zed_ros_3d
source devel/setup.bash
Follow this guide to convert a .onnx or .pt model into .engine for TensorRT (You can also use rosrun zed_ros_3d_node in this package with the same arguments and the -s flag to convert it):
If your .engine model is stored somewhere else, move it to the zed_ros_3d directory (optional)
To run the detector:
Make sure you have roscore running
roscore
Then start the node:
rosrun zed_ros_3d zed_ros_3d_node [your model].engine 0
Assuming 0 is the camera ID, it will be 0 if you only have one zed camera connected.
visualization_msgs/MarkerArray messages should now be publishing in the topic /zed_detections_raw
To view the markers, open rviz:
rosrun rviz rviz
Change the fixed frame to 'base_link', then add the MarkerArray topic