J. M. Buades

Hands and face tracking for VR applications

In this paper, we present a robust real-time 3D tracking system of human hands and face. This system can be used as a perceptual interface for virtual reality activities in a workbench environment. The main advantage of our system is that the human, placed in front of the virtual reality device, does not need any type of marker or special suit. The system includes a colour segmentation module to detect in real-time the skin-colour pixels present in the images. The results of this skin-colour segmentation will be skin-colour blobs, these are the inputs of a data association module. This module labels the blobs pixels using a set of hypothesis from previous frames. The 2D-tracking results are used for the 3D reconstruction of hands and face in order to obtain the 3D positions of these limbs. Finally, we present several results using the H-ANIM standard to show the systems output performance.

Real Time Segmentation and Tracking of Face and Hands in VR Applications

We describe a robust real-time 3D tracking system of the extreme limbs of the upper human body, i.e., the hands and the face. The goal of the system is that it can be used as a perceptual interface for virtual reality activities in a workbench environment. The whole system includes an input capture and calibration module, a real time color segmentation module, a data association and tracking module and finally a visualization VRML and H-ANIM procedure. The results of our probabilistically skin-color segmentation are skin-color blobs. Then, for each frame of the sequence our algorithm labels the blobs’ pixels using a set of object state hypothesis. This set of hypothesis is built from the results of previous frames. The 2D tracking results are used for the 3D reconstruction of limbs position in order to obtain the H-ANIM visualization results. Several results are presented to show the algorithm performance.

Matching a Human Walking Sequence with a VRML Synthetic Model

In this paper we present a specific matching technique based on basic motor patterns between two image sequences taken from different view points and a VRML synthetic human model. The criteria used are part of a generic system for the analysis and synthesis of human motion. The system has two phases: an automatic or interactively supervised analysis phase where the motion is interpreted and a synthesis phase where the original motion is applied to a biomechanical model. The main goal of our approach consists of finding a general solution that could be applied to general motor patterns. We define a set of matching conditions and we describe general-purpose criteria in order to reduce the space of search. The complexity of human motion analysis has led researchers to study new approaches and design more robust techniques for human tracking, detection or reconstruction. Whereas mathematical solutions partially solve this problem, the complexity of the algorithms proposed only serve to limit these solutions for real time purposes or general kind of motion types considered. So, we propose more simple, less general approaches but with a low computational cost. In this case the human model information about the kind of movement to be studied is very important in the process of matching between key-frame images. We also try to develop a system that can, at least in part, overcome the limitations of view dependence.

Human Body Analysis with Biomechanics Criteria

Today in many applications the study of human movement using a computer vision and graphics techniques is very useful. One of these applications is the three-dimensional reconstruction of the structure of the human body and its movement using sequences of images and biomechanical graphics models. In this paper we present a whole and general system to study the human motion without markers but, in this case, we apply it to high-level competition. This kind of study needs special procedures to do the analysis and reconstruction of the person’s body, therefore the virtual human (avatar) must have similar anthropometrical characteristics than the person who is doing the movement. We define a process to adjust the humanoid to the morphology of the person. It could be very laborious and subjective if done manually or by selection of points, but in this article we present a global human motion system capturing, modeling and matching a semiautomatic process between the real person and the modeled humanoid or synthetic avatar. Semiautomatic means that the computers propose the best matching from previous frames and the user can accept it or not.

Evaluation of K-SVD Method in Facial Expression Recognition Based on Sparse Representation Problems

The following machine learning scheme is commonly used for the recognition of facial expressions: First, the face is detected in the image. Second, tracking techniques are applied, based on active shape models; then, from the tracking of the characteristic points, a description of the facial expression is carried out, using characterization methods based on shape and/or texture; in the case of high dimension vectors, methods of features selection are applied; and finally they are classified in one of the basic expressions. In the latest years, techniques based on sparse representation methods to classify facial expression have been successfully developed. This paper aims at evaluating these methods’ performance from the training of the representation model using K-SVD. A characterization scheme of facial expression is assessed using JAFFE y CK+ databases, with or without the use of the K-SVD method, achieving a value of 0.9755 of accuracy in the classification. The obtained results prove the feasibility in the use of this method in the facial expressions classifiers based on sparse representation.

A Fast Geometric Deformation Method to Adapt a Foot to a Platform

The main goal of this research work is to develop a new system that designs shoes that adapts exactly to the foot shape. This research is based on a biomechanical anatomical structure of the foot and of the deformable shape. The system automatically selects significant foot points. We consider several anthropometrical parts of the foot in order to apply a global deformation with different axis. Also an interpolation process is designed to combine the several parts of the foot in an efficient and accurate manner. We consider different criteria in the deformation process because the top is rigid and the sole is assumed non-rigid. The system is implemented in an optimized software version in order to control the computational cost of the deformation process. A prototype of oriented commercial Application Programming Interface (API) is developed for non specialized users of the system. The results presented evaluate the error between deformations and we validate the error of several users (foot and last). Also the low error obtained guarantees the comfort of the foot that is a very important objective in this area of research.