Search

One of the most widely used motion capture (MoCap) methods depends on retroreflective markers placed on an object. Through cameras, the MoCap system captures the moving prop's (3D object) motion. However, noise in MoCap data caused by the shape of the captured volume, motion between frames, ghost points, and markers' self-occlusion could impact the quality of the captured data. To improve the quality of capturing the motion of props, we tackle the problem of finding an optimal marker set configuration for a given input prop while considering various constraints. By "props," we mean any objects or handheld items of different shapes and sizes on which markers can be precisely placed to reduce MoCap errors. We propose an approach to optimize the placement of optical (retroreflective) markers over props while encountering various constraints, such as the visibility of markers, the number of makers used, the symmetry of the marker set, and markers' overlapping. We solve the marker set configuration problem using an optimization-based method, the reversible-jump Markov chain Monte Carlo. We provide marker set configurations for various props and constraints obtained through several simulations we ran to evaluate the performance of our method. [Display omitted] • We propose a method that finds optimal marker set configurations for an input prop. • We define a total cost function that encodes various constraints as cost terms. • Our method can provide optimal marker set configurations for 3D props. • We measured our method's performance based on the total optimization time. • We conducted several evaluations of our method to understand its performance. [ABSTRACT FROM AUTHOR]

Considering the problem of recognizing non-verbal cues in Human–Robot Interaction applications, this paper proposes a novel real-time unsupervised spike timing neural network for recognition and early detection of spatio-temporal human gestures. Two spiking network classifiers one based on Izhikevich neuron model, and the other one based on Integrate-and-Fire-or-Burst neuron model have been implemented in CUDA, and allow the classification to be performed in real-time. To evaluate the performance of this proposal, we test the case of a physical robot observing air-handwritings of human gesture. The proposed approaches run in real-time, thus they are suitable for human–robot applications; they allow real-time early classifying human gestures and actions while they require a very small number of training samples. In comparing to other prominent techniques, our approaches demonstrate superior accuracy and are suitable for early classification of different types of human actions in time-sensitive mobile applications such as robotics. [ABSTRACT FROM AUTHOR]