Usage

Calibration

To calibrate the application make sure you are in the project root folder in one of the pis and run

python calibrate.py

To get (more) console output the -v flag is needed. Together with a hoop-search -o, a ball-search -b and a white calibration -w the full command which you most of the time want to use is

python calibrate.py -v -o -b -w

You can also specify the hsv color (only for the hoop so far) via the other flags. To see all of the available flags use:

python calibrate.py --help

If the -v flag is set, there will be debug pictures dumped to storage/calibration

White-calibration

Usually a white calibration is necessary everytime the lights change. This can also be introduced through different sunlight intensity.

The white calibration is tries to messure the green and blue gains to finetune the color values. Right now the white calibration takes a picture with arbitrary gains, takes a region of interest in the middle of the image (cropping it) and averaging all red, blue and green values.

For best results the ROI of the picture should be very white, or at least grey. The white calibration tries to adjust the gains that way, that it measures a gray image, instead of one with a one color overshoot.

For better convergence a p-controller is used to calculate the new gains to try.

Troubleshooting

Hoop or ball not found

If the hoop is not found have a look at the debug pictures raw-rgb.png, raw-hsv.png, hoop-mask.png and the hoop-colsplit directories and make sure the masks are clean enough and the hsv values are set correctly.

See src.ballandhoop.image.Image.color_split() for more information.

Note

The HSV values are taken as openCV interprets them (H: 0-180, S and V: 0-255 range), other applications might use different ranges to convert rgb to hsv e.g. gimp. Most common different one is H: 0-360, S and V: 0-1 or 0-100 range)

If you messure the ‘rgb’ values inside raw-hsv.png you do not have to convert the color. It is the same way as opencv/python expects it.

There are markers in the mask, but the hoop is not found

Have a look at the min_radius attributes in the config file and ajust them

I have too much noice in the hoop search / wrong hoop found

Try to limiting your HSV range more, or increase the morph_iteratrions (opening operation) attribute in config file

The ball area mask is too split up

Try to increase the morph_iterations (closing operation) in config file - it will try to connect the areas together

Camera out of memory error

Double check if the camera is connected to the rpis and the interface is activated in legacy mode.

Run application

To run the application make sure you are in the project root folder in one of the pis and run

python runner.py

To get (more) console output the -v flag is needed.

You can also specify the ball color via the other flags. To see all of the available flags use:

python runner.py --help

If the -v flag is set, there will be debug pictures dumped to storage/debug/ from every 30th frame.

You can terminate the application through pressing Ctrl+C multiple times. After the application is closed result files are written to storage/result.mat and storage/result.yml for easier plotting of the results (independent on -v flag).

Run application on all 3 pis

Do all the above steps for all 3 pis. Then first run the pi which has the serial connection and has the is_server flag.

Afterwards use the runner on the other 2 pis. If the clients disconnect at some point you can close the application and restart it. They can reconnect.