darknet_ros

Overview

A package that combines darknet with Kinect to obtain the 3D coordinates of objects.

Install

$ git clone --recursive https://github.com/rionehome/darknet_ros.git
$ cd ../../
$ catkin_make

freenect and the PointCloud Library (PCL) are required to obtain 3D information.

To install PCL, you can use the following command:

$ sudo apt-get update
$ sudo apt-get install libpcl-all

Usage

roslaunch freenect_launch freenect.launch
roslaunch darknet_ros darknet_ros_kinect.launch  

To use custom-trained weights, some configuration is necessary:

1. Add the configuration file to the cfg directory and the weights to the weights directory in ./darknet_ros/yolo_network_config.
2. Add a new YAML file to ./darknet_ros/config, following the format of existing YAML files.
3. Finally, modify and save the line <arg name="network_param_file" default="$(find darknet\_ros)/config/yolov2.yaml"/> in ./darknet_ros/launch/darknet_ros_kinect.launch to point to your newly created YAML file.

Node

name darknet_kinect
A node that sends 3D information.

Subscribe Topic

• /camera/depth_registered/points
  3D information from freenect (sensor_msgs::PointCloud2).

• /darknet_ros/bounding_boxes
  Planar coordinate information from darknet_ros (darknet_ros_msgs::BoundingBoxes).

Publish Topic

• /darknet/rgb/image_raw
  Color image converted from freenect's point cloud (sensor_msgs::Image).

• /darknet_ros/object_position
  3D coordinate information combining darknet_ros and freenect's point cloud (darknet_ros_kinect::ObjectPositions).