Catálogo de publicaciones - libros

Applying advanced video technology to understand human activity and intent is becoming increasingly important for video surveillance. In this paper, we perform automatic activity recognition by classification of spatial temporal features from video sequence. We propose to incorporate class labels information to find optimal heating time for dimensionality reduction using diffusion via random walks. We perform experiments on real data, and compare the proposed method with existing random walk diffusion map method and dual root minimal spanning tree diffusion method. Experimental results show that our proposed method is better.

With the rapid development of sensor and modeling technologies, it becomes increasingly feasible to model a large-scale environment. However, the acquisition and updating of textures for such a large-scale environment is still a challenging task, often demanding tedious and time-consuming manual interactions. This paper presents new techniques to generate high quality textures for given rough urban building models by automatic camera calibration and pose recovery, and to continuously update these textures in real time using videos as a texture resource. A number of static textures are generated for a university campus size model, and these textures are dynamically updated using videos in real time, which demonstrate the effectiveness of our algorithms.

Planar surfaces are important characteristics in man-made environments and have been successfully applied to camera calibration and interactive modeling. We develop a method for detecting planes in image pairs under epipolar constraints using planar homographies. In order to extract the whole planes, the normalized cut method is used to segment the original images. We pick those segmented regions that best fit a triangulation of the homography inliers as the detected planes. We illustrate the algorithm’s performance using gray-level and color image pairs.

A new error-resilient JPEG2000 wireless transmission scheme is proposed. The proposed scheme exploits the ‘progressive by quality’ structure of the JPEG2000 code-stream and takes into account the effect of channel errors at different quality layers in order to protect the coded bit-stream according to channel conditions using multi-rate low-density parity-check (LDPC) codes, leading to a flexible joint source-channel coding design. The novelty of this adaptive technique lies in its ability to truncate the less important source layers to accommodate optimal channel protection to more important ones to maximize received image quality. Results show that the proposed scheme facilitates considerable gains in terms of subjective and objective quality as well as decoding probability of the retrieved images.

Detecting regions of interest in video sequences is one of the most important tasks in many high level video processing applications. In this paper a novel approach based on support vector data description is presented, which detects foreground regions in videos with quasi-stationary backgrounds. The main contribution of this paper is the novelty detection approach which automatically segments video frames into background/foreground regions. By using support vector data description for each pixel, the decision boundary for the background class is modeled without the need to statistically model its probability density function. The proposed method is able to achieve very accurate foreground region detection rates even in very low contrast video sequences, and in the presence of quasi-stationary backgrounds. As opposed to many statistical background modeling approaches, the only critical parameter that needs to be adjusted in our method is the number of background training frames.

We present a novel Procedural Image Processing (PIP) method and demonstrate its applications in visualization. PIP modulates the sampling positions of a conventional image processing kernel (e.g. edge detection filter) through a procedural perturbation function. When properly designed, PIP can produce a variety of styles for edge depiction, varying on width, solidity, and pattern, etc. In addition to producing artistic stylization, in this paper we demonstrate that PIP can be employed to achieve various visualization tasks, such as contour enhancement, focus+context visualization, importance driven visualization and uncertainty visualization.

PIP produces unique effects that often either cannot be easily achieved through conventional filters or would require multiple pass filtering. PIP perturbation functions are either defined by analytical expressions or encoded in pre-generated images. We leverage the programmable fragment shader of the current graphics hardware for achieving the operations in real-time.

In this paper we present an approach for automatically detecting and tracking humans in very long video sequences. The detection is based on background subtraction using a multi-mode Codeword method. We enhance this method both in terms of representation and in terms of automatically updating the background allowing for handling gradual and rapid changes. Tracking is conducted by building appearance-based models and matching these over time. Tests show promising detection and tracking results in a ten hour video sequence.

The general goal of our research is the creation of a natural and intuitive interface for input and recognition of American Sign Language (ASL) math signs. The specific objective of this work is the development of two new interfaces for the Mathsigner application. Mathsigner is an interactive, 3D animation-based game designed to increase the mathematical skills of deaf children. The program makes use of standard input devices such as mouse and keyboard. In this paper we show a significant extension of the application by proposing two new user interfaces: (1) a glove-based interface, and (2) an interface based on the use of a specialized keyboard. So far, the interfaces allow for real-time input and recognition of the ASL numbers zero to twenty.

Existing methods of gait recognition are mostly based on either holistic shape information or kinematics features. Both of them are very important cues in human gait recognition. In this paper we propose a novel method via fusing shape and motion features. Firstly, the binary silhouette of a walking person is detected from each frame of the monocular image sequences. Then the static shape is represented using the ratio of the body’s height to width and the pixel number of silhouette. Meanwhile, a 2D stick figure model and trajectory-based kinematics features are extracted from the image sequences for describing and analyzing the gait motion. Next, we discuss two fusion strategies relevant to the above mentioned feature sets: feature level fusion and decision level fusion. Finally, a similarity measurement based on the gait cycles and two different classifiers (Nearest Neighbor and KNN) are carried out to recognize different subjects. Experimental results on UCSD and CMU databases demonstrate the feasibility of the proposed algorithm and show that fusion can be an effective strategy to improve the recognition performance.

The performance of appearance based face recognition algorithms is adversely affected by illumination variations. Illumination normalization can greatly improve their performance. We present a novel algorithm for illumination normalization of color face images. Face Albedo is estimated from a single color face image and its co-registered 3D image (pointcloud). Unlike existing approaches, our algorithm takes into account both Lambertian and specular reflections as well as attached and cast shadows. Moreover, our algorithm is invariant to facial pose and expression and can effectively handle the case of multiple extended light sources. The approach is based on Phong’s lighting model. The parameters of the Phong’s model and the number, direction and intensities of the dominant light sources are automatically estimated. Specularities in the face image are used to estimate the directions of the dominant light sources. Next, the 3D face model is ray-casted to find the shadows of every light source. The intensities of the light sources and the parameters of the lighting model are estimated by fitting Phong’s model onto the skin data of the face. Experiments were performed on the challenging FRGC v2.0 data and satisfactory results were achieved (the mean fitting error was 6.3% of the maximum color value).