Seungdo Jeong

An effective method for approximating the euclidean distance in high-dimensional space

It is crucial to compute the Euclidean distance between two vectors efficiently in high-dimensional space for multimedia information retrieval. We propose an effective method for approximating the Euclidean distance between two high-dimensional vectors. For this approximation, a previous method, which simply employs norms of two vectors, has been proposed. This method, however, ignores the angle between two vectors in approximation, and thus suffers from large approximation errors. Our method introduces an additional vector called a reference vector for estimating the angle between the two vectors, and approximates the Euclidean distance accurately by using the estimated angle. This makes the approximation errors reduced significantly compared with the previous method. Also, we formally prove that the value approximated by our method is always smaller than the actual Euclidean distance. This implies that our method does not incur any false dismissal in multimedia information retrieval. Finally, we verify the superiority of the proposed method via performance evaluation with extensive experiments.

Dimensionality reduction for similarity search with the Euclidean distance in high-dimensional applications

In multimedia information retrieval, multimedia data are represented as vectors in high-dimensional space. To search these vectors efficiently, a variety of indexing methods have been proposed. However, the performance of these indexing methods degrades dramatically with increasing dimensionality, which is known as the dimensionality curse. To resolve the dimensionality curse, dimensionality reduction methods have been proposed. They map feature vectors in high-dimensional space into vectors in low-dimensional space before the data are indexed. This paper proposes a novel method for dimensionality reduction based on a function that approximates the Euclidean distance based on the norm and angle components of a vector. First, we identify the causes of, and discuss basic solutions to, errors in angle approximation during the approximation of the Euclidean distance. Then, this paper propose a new method for dimensionality reduction that extracts a set of subvectors from a feature vector and maintains only the norm and the approximated angle for every subvector. The selection of a good reference vector is crucial for accurate approximation of the angle component. We present criteria for being a good reference vector, and propose a method that chooses a good reference vector. Also, we define a novel distance function using the norm and angle components, and formally prove that the distance function consistently lower-bounds the Euclidean distance. This implies information retrieval with this function does not incur any false dismissals. Finally, the superiority of the proposed approach is verified via extensive experiments with synthetic and real-life data sets.

Cognitive representation and Bayeisan model of spatial object contexts for robot localization

This paper proposes a cognitive representation and Bayesian model for spatial relations among objects that can be constructed with perception data acquired by a single consumer-grade camera. We first suggest a cognitive representation to be shared by humans and robots consisting of perceived objects and their spatial relations. We then develop Bayesian models to support our cognitive representation with which the location of a robot can be estimated sufficiently well to allow the robot to navigate in an indoor environment. Based on extensive localization experiments in an indoor environment, we show that our cognitive representation is valid in the sense that the localization accuracy improves whenever new objects and their spatial relations are detected and instantiated.

Virtual Tactical Map with Tangible Augmented Reality Interface

We propose a new augmented reality (AR)-based approach for training using virtual sand table representations of military battlefields. A virtual tactical map (VTM) can archive simple actions such as moving a marker by hand, to provide more organic realizations in virtual military training. The new tangible AR interface provides a content-authoring tool that is natural, intuitive, and user-friendly. AR-interfaced VTMs exhibit multiple possibilities for military learning and training applications.

News video retrieval using automatic indexing of korean closed-caption

Knowledge-based video retrieval is able to provide the retrieval result that corresponds with conceptual demand of user because of performing automatic indexing with audio-visual data, closed-caption, and so on. In this paper, we present the automatic indexing method of Korean closed-caption for knowledge-based video retrieval and the retrieval scheme using the indexed database. In the experiment, we have applied the proposed method to news video with the closed-caption generated by Korean stenographic system, and have empirically confirmed that the proposed method could provide the retrieval result that corresponds with more meaningful conceptual demand of user.

Map Representation for Robots

Map-building and localization for robots are the most basic technology required to create autonomous mobile robots. Unfortunately, they are difficult problems to comprehensively handle. If expensive sensors or a variety of external devices are used, then the problems can be resolved. However, there are still limits for various environments or platforms. Therefore, many researchers have proposed various different methods over a long period of time, and continue to do so today. In this paper, we first look at the state of existing research for map representations used in map-building and localization. We divide them into four main categories and compare the differences between them. These identified properties between the four categories can be used as good standards for choosing appropriate sensors or mathematical models when creating map-building and localization applications for robots.

Canonical View Synthesis for Gait Recognition

GAIT is defined as a persons manner of walking. While the computer vision community has progressed the identification of individuals by means of their gait, there are many difficulties which are not settled yet. Especially, dependency to the direction of walking is very serious. That is, extracted gait features become individualized according to the direction of walking, even if the camera is fixed. In this paper we propose an improvement to reduce this dependency based on homography. We establish a direction of walking through the silhouette of gait then estimate the planar homography by means of simple operations. Through transformation using planar homography we reduce dependency to the direction of the gait features. In order to estimate planar homography, vanishing point and vanishing lines are considered. The proposed planar homography that transforms the vanishing point to infinity makes the corresponding directional vanishing lines parallel, and this can always transform any plane that is established by the integrated silhouette of gait to the canonical plane. Therefore one can obtain the canonical-viewed images using the estimated homography. Moreover, through synthesizing of canonical-viewed images by means of the estimated homography, a viewpoint variation from the direction of walking is compensated. To test this idea we segment the gait silhouette into sub-regions and use the averaged features and variations on each region, as input to a gait recognition experiment. Our experiments show that the proposed method efficiently reduces the dependency to directional variations of gait.

Design of a simultaneous mobile robot localization and spatial context recognition system

In this work, we propose a simultaneous mobile robot localization and spatial context recognition system. The Harris corner detector and pyramid Lucas-Kanade optical flow are combined for robot localization. And, SIFT keypoints and its descriptors for the model-based object recognition and stereo vision technique are applied to spatial context recognition. The effectiveness of our proposed method is verified by experiments.

A framework for online gait recognition based on multilinear tensor analysis

The gait recognition is to recognize an individual based on the characteristics extracted from the gait image sequence. There are many researches for the gait recognition which use diverse kinds of information such as shape of gait silhouette, motion variation caused by walking, and so on. In general, shape information is more useful for recognition. However, shape information is influenced by a variety of factors, which degrade the recognition performance. Moreover, the information used in most of those studies might be able to be extracted after all of one or more sequences of the gait cycle are known. And it is also hard to discriminate the gait cycle from given gait sequences exactly by the online approach. In regard to these difficulties, we propose a novel gait recognition method based on the multilinear tensor analysis. To recognize the cyclic characteristic of gait without an exact division for the gait cycle, this paper’s propose is the method to form the accumulated silhouette and then describes those as the tensor. For the accumulated silhouette proposed by this paper, the image sequence of one gait cycle is divided into four sections in the training phase. However, discrimination for the gait cycle in the training phase is not directly related to the recognition phase, thus the online approach is possible. We first form the accumulated silhouettes for every individual using gait silhouettes within each section. And then, we represent these accumulated silhouettes as the tensor. Using a multilinear tensor analysis, we compute the core tensor which governs the interaction between factors organizing the original tensor, and then compose the basis to recognize the individual in the online recognition framework. Finally, we recognize the individual using the computation of similarity based on the Euclidean distance, which is more suitable to our method. We verify the superiority of the proposed approach via experiments with real gait sequences.

Dimensionality reduction in high-dimensional space for multimedia information retrieval

This paper proposes a novel method for dimensionality reduction based on a function approximating the Euclidean distance, which makes use of the norm and angle components of a vector. First, we identify the causes of errors in angle estimation for approximating the Euclidean distance, and discuss basic solutions to reduce those errors. Then, we propose a new method for dimensionality reduction that composes a set of subvectors from a feature vector and maintains only the norm and the estimated angle for every subvector. The selection of a good reference vector is important for accurate estimation of the angle component. We present criteria for being a good reference vector, and propose a method that chooses a good reference vector by using the Levenberg-Marquardt algorithm. Also, we define a novel distance function, and formally prove that the distance function consistently lower-bounds the Euclidean distance. This implies that our approach does not incur any false dismissals in reducing the dimensionality. Finally, we verify the superiority of the proposed approach via performance evaluation with extensive experiments.