A Markerless Joint Detection through a Hand Geometric Representation

Abstract

In several fields, such as man-machine interface and occupational therapy, the human hand-joint position is required. Traditional methods usually rely on image processing allied with marker placement, e.g. reflexive marker, which can be time-consuming and uncomfortable for the subject. For these reasons, scientific efforts are being made to create reliable and convenient joint tracking. This paper proposes a methodology that generates geometric figures to mimic the hand configuration. This process is made possible by an optimization algorithm, which finds the most suitable placement of these geometric shapes. One time the real hand and the created representation share similar features, the joints position can be estimated. Two optimization algorithms were employed: particle swarm optimization and genetic algorithm. In both cases, satisfactory results were obtained. Although, particle swarm optimization marginally outperformed the latter method.