stronghold-vision-opencv

Vision process with OpenCV by Zodiac.

License

MIT

声明：我们允许组织以及个人将此项目用于商业行为，没有任何限制。 但请带上本项目的许可证 MIT，并附上原作者以及项目的地址。我们保留对该项目的最终解释权。

Installation

Python3, numpy, cv2, matplotlib are required to run this script. To communicate with roboRIO, you also need pynetworktables.

Install OpenCV 3.1 for Python3, and then install other requirements:

git clone https://github.com/ZodiacEFZ/stronghold-vision-opencv && cd stronghold-vision-opencv

pip install -r requirements.txt

Analyze

Put all test images in ./TestData/RealFullField folder, then run ./analyze.py. Then there will be a window showing analyze result.

Uncomment cv2.imwrite("TestData/Result/res%d.png" % (i), img) then result images will be put in ./TestData/Result with contours drawn on.

Production

Run ./video.py, then OpenCV will read data from the default camera and analyze data. With NetworkTables you can enable/disable vision process, manual exposure. And this vision daemon will report target data to NetworkTables. Also an mjpeg server will be set up to transfer processed image stream.

Structure

detect folder stores detect algorithm.

grip folder stores origin grip solution.

Config

Create config.py. Here's an example.

server = {
    "host": "",
    "port": 8080,
    "rioAddress": "roboRIO-9036-FRC.local"
}
camera = {
    "width": 640,
    "height": 480,
    "fps": 30,
    "r_width": 322,
    "r_height": 245,
    "cX": 320,
    "cY": 240,
    "id": 0
}

How to use

After running ./video.py, vision daemon will start an http server serving mjpeg stream at specified port. Because this server only support one connection, You should use a mjpeg proxy to transfer video data.

If you want to view vision settings, deploy a WebDashboard.