Techniques for Real-time Multi-person Face Tracking for Human-robot Dialogue

Abstract

The aim of this work is an investigation of interaction between a robot and multiple humans. The head robot is “FurHat” - in Speech, Music and Hearing (TMH) department in KTH - and the most focus is on real-time tracking algorithms. The study contains two parts: first, selecting an efficient algorithm by comparing some of the market’s offers in detecting and/or tracking area; second, establishment of a method which has the possibility of starting a real-time interaction between FurHat and multi-person. A machine learning algorithm in Matlab suggests a method to calculate the coefficients for “Priority Function” which is implemented on the outputs of SHORE library. This function can guide the robot to select the target face which should be chosen for tracking or starting a conversation. The project is done at KTH in TMH department.