Recently, YOLOv5 (although there seems to be pros and cons) came out, so I decided to change from YOLOv3 to YOLOv4, which is a little outdated, but I could not find a setup article in the ROS (Robot Operating System) environment. I will summarize it as a memorandum.
Summary of each version of YOLO (Qiita)
This time, we confirmed the operation on Ubuntu 16.04 and Ubuntu 18.04. Details are below.
| OS | Ubuntu16.04.6 LTS | Ubuntu18.04.5 LTS |
|---|---|---|
| CPU | Intel Core i7-8700 | Intel Core i5-6500 |
| GPU | Geforce GTX 1080 Ti | Geforce GTX 960 |
| ROS | Kinetic 1.12.14 | Merodic 1.14.9 |
| Nvidia Driver | 450.51.06 | 440.100 |
| CUDA(toolkit) | 8.0 | 10.2 |
| cuDNN | 7 | 7 |
This time, we will proceed on the premise that the above environment is in place.
Currently, evacuate the ** darknet_ros package ** under ** ~ / catkin_ws / src / **. Create a backup directory (YOLOv3 this time) in your home directory and move it there.
Terminal
#Please execute line by line
$ cd
$ mkdir YOLOv3
$ cd ~/catkin_ws/src
$ mv darknet_ros ~/YOLOv3
Next, clone the ** darknet_ros ** package for YOLO v4 by referring to the following.
YOLO V4 for darknet_ros
Terminal
$ cd ~/catkin_ws/src
$ git clone --recursive https://github.com/Tossy0423/darknet_ros.git
$ cd ..
$ catkin_make
If make passes, you're done.
This ** darknet_ros uses ** uvc_camera ** as the driver for the ** USB camera. In my case, I'm using ** usb_cam ** as a driver, so I had to change the camera node that darknet received when testing.
Change before
1 subscribers:
2
3 camera_reading:
4 topic: /camera/rgb/image_row
5 queue_size: 1
After change
1 subscribers:
2
3 camera_reading:
4 topic: /usb_cam/image_row
5 queue_size: 1
I tried an operation test.
Terminal
#Run on two terminals
# darknet_ros node launch
$ roslaunch darknet_ros yolov4.launch
#Launch camera node
$ roslaunch usb_cam usb_cam-test.launch
I was able to display it nicely. that's all.
Recommended Posts
