yolov5-ros1

Simple yolov5 ros1 wrapper.

Subscribes the image topic and inference by yolov5 to get bounding box and annotations.

You can use webcam as an image publisher.

1) Download this repository to workspace/src

cd catckin_ws/src
git clone --recursive https://github.com/jellyho/yolov5-ros1.git

2) Install python dependencies of yolov5

cd yolov5-ros1
cd yolov5
pip install -r requirements.txt

3) Put pretrained weights

copy-paste the pretrained weights into catkin_ws/src/yolov5/src

4) Build the package and run!

cd ~/catkin_ws
catkin build
source devel/setup.bash

There are three options when execute launch file.

• image(string) : Image topic name that you want to apply yolov5.
• verbose(bool) : Open popup window that shows the annotated results.
• publish(bool) : Publish the annotated image
• weights(stirng) : Pretrained weight name in src/ folder

Default setting is verbose:=false, publish:=true

If you don't put any option about weights, defuault yolov5n.pt will applied.

.

1. Subscribe existing image topic and yolo

roslaunch yolov5 yolo.launch image:='/topic_name' verbose:=false publish:= true weights:=yolov5m.pt

2. Use webcam as an image publisher

roslaunch yolov5 yolo_webcam.launch vebose:=false publish:=true

5) Topic lists

/yolo_image - Image - Annotated inferenced image when publish:=true
/yolo_results - String - Bounding box information encoded into json format

How to use /yolo_results ?

# define String msg subscriber
rospy.Subscriber('/yolo_results', String, yolo_cb)

# In callback, convert to json
import json

def yolo_cb(msg):
    result = json.loads(msg.data)
    print(result)

[{'xmin': 140.4574737549, 'ymin': 1.7290496826, 'xmax': 509.8947143555, 'ymax': 480.0, 'confidence': 0.5377990603, 'class': 0, 'name': 'person'},
 {'xmin': 25.5067329407, 'ymin': 366.9152832031, 'xmax': 146.5026702881, 'ymax': 455.7973632812, 'confidence': 0.520999074, 'class': 64, 'name': 'mouse'}]