darknet_ros

Why we included this project

Robot developers who already build on ROS and want real-time object detection without bolting on a separate pipeline will get the most out of this package. It wraps the YOLOv3 network so camera frames come in and bounding-box detections go out as regular ROS messages, which downstream nodes can subscribe to like any other topic. Pre-trained models for the VOC and COCO class sets are included, so common objects such as people, vehicles, and animals are detected right away, and you can retrain for your own categories. Setup is heavier than a pip-install Python library, since you need a ROS distribution and the Darknet build working first, but for a ROS stack that is the point: the detector behaves like part of the system rather than a separate process. It is research code from ETH Zurich, so expect rough edges and treat it as a prototyping base rather than a production product.