history-vector-phase-detection

About 200 coronary angiograms were recorded from a distance in a cardiac catheterisation laboratory at the Reinier de Graaf Gasthuis, Delft, NL.

The purpose of the video recordings was to analyse workflow during procedures.

This Python repository aims to recognise workflow phases from human motion.

It analyses bodypart motion, human posture, and positioning with respect to other persons.

After encoding all these aspect into history vectors, it builds a mixture model for classification of new motions.

Additionally, it takes procedure duration into account.

Unfortunately, this approach proved unable to accurately classify workflow phases, so using this code for that purpose is not recommended.