However, this pose was not captured for the dancers and therefore custom skeleton
tracking calibration for each dancer is not possible. Fortunately, our experiments show that skeleton tracking is quite effective even if non-exact calibration data is provided to
the OpenNI tracking module. In this work, we created custom calibration results for each dancer by manually sourcing and calibrating persons with similar body characteristics to
each dancer in the dataset. This tailored calibration data
resulted in more robust skeletal tracking and consequently
higher accuracy of the automatic evaluation methods.