Eraldo Ribeiro

Human motion recognition using Isomap and dynamic time warping

In this paper, we address the problem of recognizing human motion from videos. Human motion recognition is a challenging computer vision problem. In the past ten years, a number of successful approaches based on nonlinear manifold learning have been proposed. However, little attention has been given to the use of isometric feature mapping (Isomap) for human motion recognition. Our contribution in this paper is twofold. First, we demonstrate the applicability of Isomap for dimensionality reduction in human motion recognition. Secondly, we show how an adapted dynamic time warping algorithm (DTW) can be successfully used for matching motion patterns of embedded manifolds. We compare our method to previous works on human motion recognition. Evaluation is performed utilizing an established baseline data set from the web for direct comparison. Finally, our results show that our Isomap-DTW method performs very well for human motion recognition.

Recognizing primitive interactions by exploring actor-object states

In this paper, we present a solution to the novel problem of recognizing primitive actor-object interactions from videos. Here, we introduce the concept of actor-object states. Our method is based on the observation that at the moment of physical contact, both the motion and the appearance of actors are constrained by the target object. We propose a probabilistic framework that automatically learns models in such constrained states. We use joint probability distributions to represent both actor and object appearances as well as their intrinsic spatio-temporal configurations. Finally, we demonstrate the applicability of our approach on series of human-object interaction classification experiments.

Laboratory screening of coating libraries for algal adhesion.

Abstract Coatings libraries achieved through a combinatorial chemistry approach, which may generate tens to hundreds of formulations, can be deposited in an array of 12 patches, each approximately 9 cm2, on 10 × 20 cm primed aluminum panels. However, existing methods to quantify algal biomass on coatings are unsuitable for this type of array format. This paper describes an algorithm modelled on a probability distribution that quantifies the area of surface covered by a green alga from digital images. The method allows coatings with potential fouling-release properties to be down-selected for further evaluation. The use of the algorithm is illustrated by a set of eight siloxane-polyurethane coatings made using organofunctional poly(dimethylsiloxane) (PDMS) and poly(ϵ-caprolactone)-PDMS-poly(ϵ-caprolactone) (PCL-PDMS-PCL) triblock copolymers along with four PDMS standards which were deposited on one panel. Six replicate panels were seeded with Ulva zoospores which grew into sporelings (small plants) that completely covered the surface. The ease of removal of the Ulva sporeling biofilms was determined by automated water jetting at six different impact pressures. The coverage of the biofilm on the twelve individual formulations after jet washing was quantified from the green colour of digital images. The data are discussed in relation to the composition of the coatings.

Computer Vision for Nanoscale Imaging

The main goal of Nanotechnology is to analyze and understand the properties of matter at the atomic and molecular level. Computer vision is rapidly expanding into this new and exciting field of application, and considerable research efforts are currently being spent on developing new image-based characterization techniques to analyze nanoscale images. Nanoscale characterization requires algorithms to perform image analysis under extremely challenging conditions such as low signal-to-noise ratio and low resolution. To achieve this, nanotechnology researchers require imaging tools that are able to enhance images, detect objects and features, reconstruct 3D geometry, and tracking. This paper reviews current advances in computer vision and related areas applied to imaging nanoscale objects. We categorize the algorithms, describe their representative methods, and conclude with several promising directions of future investigation.

Shape from periodic texture using the eigenvectors of local affine distortion

Shows how the local slant and tilt angles of regularly textured curved surfaces can be estimated directly, without the need for iterative numerical optimization. We work in the frequency domain and measure texture distortion using the affine distortion of the pattern of spectral peaks. The key theoretical contribution is to show that the directions of the eigenvectors of the affine distortion matrices can be used to estimate local slant and tilt angles of tangent planes to curved surfaces. In particular, the leading eigenvector points in the tilt direction. Although not as geometrically transparent, the direction of the second eigenvector can be used to estimate the slant direction. The required affine distortion matrices are computed using the correspondences between spectral peaks, established on the basis of their energy ordering. We apply the method to a variety of real-world and synthetic imagery.

Scale and rotation invariant detection of singular patterns in vector flow fields

We present a method for detecting and describing features in vector flow fields. Our method models flow fields locally using a linear combination of complex monomials. These monomials form an orthogonal basis for analytic flows with respect to a correlation-based innerproduct. We investigate the invariance properties of the coefficients of the approximation polynomials under both rotation and scaling operators. We then propose a descriptor for local flow patterns, and developed a method for comparing them invariantly against rigid transformations. Additionally, we propose a SIFT-like detector that can automatically detect singular flow patterns at different scales and orientations. Promising detection results are obtained on different fluid flow data.

Improved Reconstruction of Images Distorted by Water Waves

This paper describes a new method for removing geometric distortion in images of submerged objects observed from outside shallow water. We focus on the problem of analyzing video sequences when the water surface is disturbed by waves. The water waves will affect the appearance of the individual video frames such that no single frame is completely free of geometric distortion. This suggests that, in principle, it is possible to perform a selection of a set of low distortion sub-regions from each video frame and combine them to form a single undistorted image of the observed object. The novel contribution in this paper is to use a multi-stage clustering algorithm combined with frequency domain measurements that allow us to select the best set of undistorted sub-regions of each frame in the video sequence. We evaluate the new algorithm on video sequences created both in our laboratory, as well as in natural environments. Results show that our algorithm is effective in removing distortion caused by water motion.

Estimating the 3D orientation of texture planes using local spectral analysis

This paper describes a new method for recovering the perspective geometry of textured planes from local spectral analysis. The novel contribution of the work is to use the angular distribution of texture moments to locate the vanishing point of textured planes viewed under perspective geometry. Our key observation is that lines of uniform spectral orientation radiate from the vanishing point. We exploit this property to develop a simple algorithm for estimating the slant and tilt of textured planes. This is a two-step process. It commences by computing the Fourier power spectrum at each of a series of local neighbourhoods and identifying contours which connect the local spectral moments of equal power. As noted by Sevens (1981), these contours are perpendicular to the tilt direction. The second step is to triangulate the vanishing point. We do this by using a correlation method to identify lines of uniform spectral orientation that originate from the vanishing point. This second property represents the novel contribution of the paper. We evaluate the new algorithm on both synthetic imagery with known ground-truth and on real-world data. q 2000 Elsevier Science B.V. All rights reserved.

Learning human motion models from unsegmented videos

We present a novel method for learning human motion models from unsegmented videos. We propose a unified framework that encodes spatio-temporal relationships between descriptive motion parts and the appearance of individual poses. Sparse sets of spatial and spatio-temporal features are used. The method automatically learns static pose models and spatio-temporal motion parts. Neither motion cycles nor human figures need to be segmented for learning. We test the model on a publicly available action dataset and demonstrate that our new method performs well on a number of classification tasks. We also show that classification rates are improved by increasing the number of pose models in the framework.

3-D Planar Orientation from Texture: Estimating Vanishing Point from Local Spectral Analysis.

This paper describes a novel method for recovering the perspective geometry of textured surfaces from local spectral moments. It commences by computing the Fourier power spectrum at each of a series of local neighbourhoods. We illustrate how the vanishing point of the global perspective geometry can be estimated from local spectral moments. This is a two-step process. We commence by identifying contours which connect the local spectral moments of equal power. These contours are perpendicular to the tilt direction at the image plane. The second step is to triangulate the vanishing point. We do this by using a correlation method to identify lines with identical oriented spectral moment. In this way we can use a sample of spectral moments estimate the vanishing point location.