Technical Reports of the Computer Science Department

Permanent URI for this collectionhttp://hdl.handle.net/1903/6

Browse

Author: search.filters.author.Aloimonos, Yiannis

Search Results

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    Spatial and spatio-temporal characterization of movement for the analysis of actions and actors
    (2011-10-13) Claudino, Leonardo; Aloimonos, Yiannis
    Movement data is high-dimensional but often redundant, meaning there is certainly a lower dimensional subspace that spans most of the body configurations within an action performance. We propose that one such representation can be achieved through a decomposition method that explores the existence of key configurations and temporal correlations of those configurations that are typical of action matrices. The approach is compatible with computational models of motor synergies based on matrix factorizations, and it builds upon a method that was earlier proposed in the context of biological motion perception. Our experiments show that vertical jump trials collected from children and young adults can be consistently reconstructed from the resulting representation. We also observe that a subset of that same representation suggests differences among populations of jumpers based on their trials, which serves to illustrate the potential of the method as a tool to analyze both actions and actors.
  • Thumbnail Image
    Item
    An Approximate Linear Approach for the Fundamental Matrix Computation
    (2003-10-06) Guerra-Filho, Gutemberg; Aloimonos, Yiannis
    We introduce a new robust approach for the computation of the fundamental matrix taking into account the intrinsic errors (uncertainty) involved in the discretization process. The problem is modeled as an approximate equation system and reduced to a linear programming form. This approach is able to compute the solution set instead of trying to compute only a single vertex of the solution polyhedron as in previous approaches. Therefore, our algorithm is a robust generalization of the eight-point algorithm. The exact solution computation feasibility is proved for some pure translation motions, depending only on the distribution of the discretization errors. However, a single exact solution for the fundamental matrix is not feasible for pure rotation cases. The feasibility of an exact solution is decided according to an error distance between a nontrivial exact solution and the faces of the solution set.
  • Thumbnail Image
    Item
    An Optimal Time-Space Algorithm for Dense Stereo Matching
    (2003-10-06) Guerra-Filho, Gutemberg; Aloimonos, Yiannis
    An original survey addressing time-space complexity covers several stereo matching algorithms and running time experiments are reported. Taking the point of view that good reconstruction needs to be solved in feedback loops, we then present a new dense stereo matching based on a path computation in disparity space. A procedure which improves disparity maps is also introduced as a post-processing step for any technique solving a dense stereo matching problem. Compared to other algorithms, our algorithm has optimal time-space complexity. The algorithm is faster than "real-time" techniques while producing comparable results. The correctness of our algorithm is demonstrated by experiments in real and synthetic benchmark data.
  • Thumbnail Image
    Item
    A Language for Human Action
    (2006-10-15) Guerra-Filho, Gutemberg; Aloimonos, Yiannis
    Human-centered computing (HCC) is centered on humans and what they do, i.e. human actions. Thus, developing an infrastructure for HCC requires understanding human action, at some level of detail. We need to be able to talk about actions, synthesize actions, recognize actions, manipulate actions, imitate actions, imagine and predict actions. How could we achieve this in a principled fashion? This paper proposes that the space of human actions has a linguistic structure. This is a sensory-motor space consisting of the evolution of the joint angles of the human body in movement. The space of human activity has its own phonemes, morphemes, and sentences. We present a Human Activity Language (HAL) for symbolic non-arbitrary representation of visual and motor information. In phonology, we define atomic segments (kinetemes) that are used to compose human activity. In morphology, we propose parallel learning to incorporate associative learning into a language inference approach. Parallel learning solves the problem of overgeneralization and is effective in identifying the active joints and motion patterns in a particular action. In syntax, we point out some of the basic constraints for sentence formation. Finally, we demonstrate this linguistic framework on a praxicon of 200 human actions (motion capture data obtained by a suit) and we discuss the implications of HAL on HCC.
  • Thumbnail Image
    Item
    Learning Parallel Grammar Systems for a Human Activity Language
    (2006-04-26) Guerra-Filho, Gutemberg; Aloimonos, Yiannis
    We have empirically discovered that the space of human actions has a linguistic structure. This is a sensory-motor space consisting of the evolution of the joint angles of the human body in movement. The space of human activity has its own phonemes, morphemes, and sentences. In kinetology, the phonology of human movement, we define atomic segments (kinetemes) that are used to compose human activity. In this paper, we present a morphological representation that explicitly contains the subset of actuators responsible for the activity, the synchronization rules modeling coordination among these actuators, and the motion pattern performed by each participating actuator. We model a human action with a novel formal grammar system, named Parallel Synchronous Grammar System (PSGS), adapted from Parallel Communicating Grammar Systems (PCGS). We propose a heuristic PArallel Learning (PAL) algorithm for the automatic inference of a PSGS. Our algorithm is used in the learning of human activity. Instead of a sequence of sentences, the input is a single string for each actuator in the body. The algorithm infers the components of the grammar system as a subset of actuators, a CFG grammar for the language of each component, and synchronization rules. Our framework is evaluated with synthetic data and real motion data from a large scale motion capture database containing around 200 different actions corresponding to verbs associated with voluntary observable movement. On synthetic data, our algorithm achieves 100% success rate with a noise level up to 7%.
  • Thumbnail Image
    Item
    On the Geometry of Visual Correspondence
    (1998-10-15) Fermuller, Cornelia; Aloimonos, Yiannis
    (Also cross-refernced as CAR-TR-732) Image displacement fieldsoptical flow fields, stereo disparity fields, normal flow fieldsdue to rigid motion possess a global geometric structure which is independent of the scene in view. Motion vectors of certain lengths and directions are constraine d to lie on the imaging surface at particular loci whose location and form depends solely on the 3D motion parameters. If optical flow fields or stereo disparity fields are considered, then equal vectors are shown to lie on conic sections. Similarly, for normal motion fields, equal vectors lie within regions whose boundaries also constitute conics. By studying various properties of these curves and regions and their relationships, a characterization of the structure of rigid motion fields is given. The go al of this paper is to introduce a concept underlying the global structure of image displacement fields. This concept gives rise to various constraints that could form the basis of algorithms for the recovery of visual information from multiple views.
  • Thumbnail Image
    Item
    Vision and Action
    (1998-10-15) Fermuller, Cornelia; Aloimonos, Yiannis
    (Also cross-referenced as CAR-TR-722) Our work on Active Vision has recently focused on the computational modelling of navigational tasks, where our investigations were guided by the idea of approaching vision for behavioral systems in form of modules that are directly related to perceptual tasks. These studies led us to branch in various directions and inquire into the problems that have to be addressed in order to obtain an overall understanding of perceptual systems. In this paper we present our views about the architecture of vision syst ems, about how to tackle the design and analysis of perceptual systems, and promising future research directions. Our suggested approach for understanding behavioral vision to realize the relationship of perception and action builds on two earlier approac hes, the Medusa philosophy 13] and the Synthetic approach [15 The resulting framework calls for synthesizing an artificial vision system by studying vision corr petences of increasing complexity and at the same time pursuing the integration of the percept ual components with action and learning modules. We expect that Computer Vision research in the future will progress in tight collaboration with many other disciplines that are concerned with empirical approaches to vision, i.e. the understanding of biolo gical vision. Throughout the paper we describe biological findings that motivate computational arguments which we believe will influence studies of Computer Vision in the near future.