Key Point Labelling
Key Point Labelling
As part of our efforts to ramp up for production, we require an AI model that can
detect key points of a football field. In order to this we must gather a dataset
compromised of images that the AI will likely see and label all the key points in
view in order to train the AI.
Our AI model will comprise of 32 key points of a football field. The key points will
span from the top left corner of the football field all the way down to the bottom
right corner of the field. Below is a diagram of a 2D football field showing where
each key point should be located:
As you can see from the above, the points start in the top left and work there
way downwards, moving towards the right.
To label the images in our dataset we have created a Roboflow project. A class
named “pitch” has been created to use. Within this class the key points reside.
The labellers job is to first highlight the pitch using the class, and then move all
the key points to the required location on the image. Any key points that are not
in view in the image need to be deleted.
Below are examples of how an image will look after completing a labelling task
on it:
As you can see not all 32 points have to be placed on the image, only the ones
that are in view. Some pictures in the dataset will however require all the
points are they are panorama images.