Text this: Recognition of human motion from qualitative normalised templates