A vision-based human intended action tracker using HMM and TD learning algorithm

摘要

In this paper, we propose a close-looped system with a model using visual information to track human operator, and detect his command, comply that, and at last verily reward from an operator. The model integrate the Temporal Difference (TD) Learning Algorithm with the environment model made by Hidden Markov Model (HMM) which infers the human walking model, and his intended action model. The result is a sequence of actions from the model that makes the camera successfully track and detect the intended command. We demonstrate the method in the system helps a man coming from a distance to give sign-command about the number of things to bring out of a store.