Catálogo de publicaciones - libros

This paper tackles a particular shape matching problem: given a data base of shapes (described as triangular meshes), we search for all shapes which describe a human. We do so by applying a 3D face detection approach on the mesh which consists of three steps: first, a local symmetry value is computed for each vertex. Then, the symmetry values in a certain neighborhood of each vertex are analyzed for building sharp symmetry lines. Finally, the geometry around each vertex is analyzed to get further facial features like nose and forehead. We tested our approach with several shape data bases (e.g. the Princeton Shape Benchmark) and achieved high rates of correct face detection.

Among many skin color sources, hemoglobin and melanin components are regarded most significant in determining the final color of human skin. In extracting the two components, this paper proposes to use a more intuitive color model, HSV, than the RGB color space. For 3D rendering, we propose a novel method to reconstruct the fine-scale structure (normal map) of the human skin from a single highly specular image, or from a pair of highly specular and diffusive images. In the shape-from-shading algorithm, several photos varying light positions should have been taken to compute the normal of diffusive surfaces; which, however, was hardly the case for capturing fine skin structure since the human subjects cannot keep still while taking many photos. Therefore, our approach rids us of the difficult task to register several images.

We introduce an efficient approach for representing a human face using a limited number of images. This compact representation allows for meaningful manipulation of the face. Principal Components Analysis (PCA) utilized in our research makes possible the separation of facial features so as to build statistical shape and texture models. Thus changing the model parameters can create images with different expressions and poses. By presenting newly created faces for reviewers’ marking in terms of intensities on masculinity, friendliness and attractiveness, we analyze relations between the parameters and intensities. With feature selections, we sort those parameters by their importance in deciding the three aforesaid aspects. Thus we are able to control the models and transform a new face image to be a naturally masculine, friendly or attractive one. In the PCA-based feature space, we can successfully transfer expressions from one subject onto a novel person’s face.

We present a learning-based 3D facial expression mapping technique that preserves facial expression details and works in real time for even high resolution meshes. Our approach is inspired by a previously developed technique called deformation transfer [1]. The deformation transfer technique preserves facial expression details but its computational overhead makes it not suitable for real time applications. To accelerate computation, we use a piecewise linear function to represent the mapping from the direct motion (the difference between the expression face and the neutral face) to the motion obtained by the deformation transfer method. This piecewise linear function is learned offline from a small set of training data. The online computation is thus reduced to the evaluation of the piecewise linear functions which is significantly faster. As a result, we are able to perform real time expression mapping for even high resolution 3D meshes.

This paper describes an experiment that was conducted to evaluate three interaction techniques aiming at interacting with large virtual environments using haptic devices with limited workspace: the Scaling technique, the Clutching technique, and the Bubble technique. Participants were asked to paint a virtual model as fast and as precisely as possible inside a CAVE, using a “desktop” haptic device. The results showed that the Bubble technique enabled both the quickest and the most precise paintings. It was also the most appreciated technique.

Natural grasp interaction plays an important role in enhancing users’ immersion experience in virtual environments. However, visually distracting artifacts such as the interpenetration of the hand and the grasped objects are always accompanied during grasp interaction due to a simplified whole-hand collision model, discrete control data used for detecting collisions and the interference of device noises. In addition, complicated distribution of forces from multi-finger contacts makes the natural grasp and manipulation of a virtual object difficult. In order to solve these problems, this paper presents a novel approach for grasp interaction in virtual environments. Based on the research in Neurophysiology, we first construct finger’s grasp trajectories and detect collisions between the objects and the trajectories instead of the whole-hand collision model, then deduce the grasp configuration using collision detection results, and finally compute feedback forces according to grasp identification conditions. Our approach has been verified in a CAVE-based virtual environment.

Pen-based user interface has become a hot research field in recent years. Pen gesture plays an important role in Pen-based user interfaces. But it’s difficult for UI designers to design, and for users to learn and use. In this purpose, we performed a research on user-centered design and recognition pen gestures. We performed a survey of 100 pen gestures in twelve famous pen-bases systems to find problems of pen gestures currently used. And we conducted a questionnaire to evaluate the matching degree between commands and pen gestures to discover the characteristics that a good pen gestures should have. Then cognition theories were applied to analyze the advantages of those characteristics in helping improving the learnability of pen gestures. From these, we analyzed the pen gesture recognition effect and presented some improvements on features selection in recognition algorithm of pen gestures. Finally we used a couple of psychology experiments to evaluate twelve pen gestures designed based on the research. It shows those gestures is better for user to learn and use. Research results of this paper can be used for designer as a primary principle to design pen gestures in pen-based systems.

The main challenges of realistically simulating the displacement of humanoid pedestrians are twofold: they need to behave realistically and they should accomplish their tasks. Here we present a field potential formalism, based upon boundary value problems, that allows a group of synthetic actors to move negotiating space, avoiding collisions, attaining goals in prescribed sequences while at same time producing very individual paths. The individuality of each pedestrian can be set by changing its inner field parameters. This leads to a broad range of possible behaviors without jeopardizing its task performance. Simulate situations as behavior in corridors, collision avoidance and competition for a goal are presented and discussed.

In recent years, large scale distributed virtual environments (DVEs) have become a major trend in distributed applications, mainly due to the enormous popularity of multiplayer online games in the entertainment industry. Since architectures based on networked servers seems to be not scalable enough to support massively multiplayer applications, peer-to-peer (P2P) architectures have been proposed as an efficient and truly scalable solution for this kind of systems. However, the main challenge of P2P architectures consists of providing each avatar with updated information about which other avatars are its neighbors. We have denoted this problem as the awareness problem. Although some proposals have been made, none of them provide total awareness to avatars under any situation. This paper presents a new awareness method based on unicast communication that is capable of providing awareness to 100% of avatars, regardless of both their location and their movement pattern in the virtual world. Therefore, it allows large scale DVEs based on P2P architectures to properly scale with the number of users while fully providing awareness to all of them.

In this paper, a novel motion editing tool, called the state feedback dynamic model, is proposed and demonstrated for the animators to edit the pre-existing motion capture data. The state feedback dynamic model is based on the linear time-invariant system (LTI). Compared with previous works, by this model, the animators need only modify a few keyframes manually, and the other frames can be adjusted automatically while preserving as much of the original quality as possible. It is a global modification on motion sequence. More important, the LTI model derives an explicit mapping between the high-dimensional motion capture data and low-dimensional hidden state variables. It transforms a number of possibly correlated joint angle variables into a smaller number of uncorrelated state variables. Then, the motion sequence is edited in state space, and which considers that the motion among joints is correlated. It is different from traditional methods which consider each joint as independent of each other. Finally, an effective algorithm is also developed to calculate the model parameters. Experimental results show that the generated animations through this method are natural and smooth.