Presenting a ROS 2 node for 3D object detection in point clouds using a pretrained model from NVIDIA TAO Toolkit based on PointPillars.
Accurate, fast object detection is a key task for safe robotic navigation and this node takes advantage of performing detection on lidar input over image-based systems (lidar is not sensitive to changing lighting conditions, unlike cameras). The node takes point clouds as input from real or simulated lidars, performs TensorRT-optimized inference to detect objects and outputs 3D bounding boxes as a Detection3DArray message for each point cloud.
The PointPillar model we used detects objects of three classes: Vehicle, Pedestrian, and Cyclist. You can train your own detection model following the TAO toolkit steps and use it with this node!
For more details, check out the technical blog and NVIDIA-AI-IOT/ros2_tao_pointpillars on GitHub.