Face Tracking Robot testbed for Performance Assessment of Machine Learning Techniques

Abstract

Computer vision is an essential part of robotics. Face detection and tracking has improved with improvements in machine learning techniques. Various techniques has been proposed which work well with high performance computing facilities. We assess the comparative performance of various machine learning techniques to improve face tracking and implement a robot which can respond to user behaviour. Simulations have been done prior to construction of robot to obtain smooth movement. Earlier computer vision experiments using robots used simpler image processing algorithms. Machine learning and internet connected face tracking would suffer from processing delays affecting real time behavior. Here our experimental results show performance comparison of popular face recognition techniques running on an offline computing machine. Given the accuracy benefits, we feel adopting machine learning based techniques as opposed to image processing is acceptable even in resource constrained environment.