A phase estimation method for workflow based on optical flow and Hidden Markov Model

In this paper, we proposed a method to estimate phases in a workflow. The workflow is defined as certain works that have to be done in a certain order. The goal of the proposed method is to estimate the current phases that represent the works from the recorded data of workflow. The assumed workflow in this paper is a cholecystectomy surgery that is a kind of laparoscopic surgery and the surgery can be defined as the basic order of processes. The proposed method calculates optical-flows from the recorded images and uses them as the word of Latent Dirichlet Allocation to cluster the optical-flows into patterns called topics. After that, a Hidden Markov Model was utilized to estimate the phases by the topics used as observed values. In the experiment, the several recorded data from the assumed workflow were used for estimation and the accuracy of proposed method was checked by leave one out cross validation on the data.