YOLOv3 for ROS

Overview

This is an extension of YOLO: Real-Time Object Detectionusing ROS wrapper to implement the function of YOLOv3 as a ROS node.

The node is written in Python and follows the example included in the original darknet package.

This package has been tested on Ubuntu 16.04 and ROS Kinetic.

Author: Zi-Fan WANG, [email protected]

Citing

About Darknet: https://github.com/pjreddie/darknet YOLOv3 method is original described in https://pjreddie.com/media/files/papers/YOLOv3.pdf

Result

Installation

Dependencies

• OpenCV3
• Scipy

Build

Go to your workspace and download the repo:

git clone --recursive https://github.com/SailColubrid/yolo3_ros.git

Build Darknet first, refer to https://pjreddie.com/darknet/install/

cd yolo3_ros
cd darknet

if you want to use GPU and cudnn. Please make sure you change

OPENCV = 0

in Makefile into

OPENCV = 1

to enable the use of OpenCV. And for ros_yolo can use the files in darket, you also have to change the paths in darket/cfg/coco.data:

valid = /[YOUR_PATH_TO_DARKNET]/data/coco_val_5k.list
names = /[YOUR_PATH_TO_DARKNET]/data/coco.names

Then:

make -j4

Then go to /catkin_ws/src/yolo3_ros/src/yolo_node.py, go to line 64 and change the path to .so file to yours.:

lib = CDLL("/[YOUR_PATH_TO_DARKENT]/libdarknet.so", RTLD_GLOBAL)

Build the ROS package

cd ~/catkin_ws  # I assume the name of your workspace is catkin_ws
catkin_make
source devel/setup.bash

Configuration

• yolo_node subscribes to: /camera_raw and publishes /detection. You can use rviz to display the result.

Usage

Provided Example: Using web camera

  roslaunch yolo3_ros demo_web.launch