Sangkyoon Kim

Human Action Recognition Using Product Manifold Theory and Procrustes Shape Analysis

The shape space that is composed of the configurations matrices of landmarks extracted from a given video is represented by a product manifold. Therefore, we proposed new method that can recognize a human action using mathematical properties of product manifold and Procrustes shape analysis. In the first step, we factorize each volume images belonging to training dataset as a time ordering sequence of images, and we extract pre-shape configuration vector of landmarks from each frames consisting an image sequence. Then, we have obtained a random sample of pre-shape configuration vectors from all videos stored in tanning database by using similar procedure, and we compute mean shape vectors for random sample of extracted shape vectors. In the second step, in order to recognize the query human action video, we derive a sequence of the pre-shape configuration vectors from given query video, and we project each shape vector on the tangent space with respect to the pole taking on a sequence of the mean shape vectors corresponding with a target video. We recognize a query video as target video that can minimize the distance between two sequences of the pre-shape vectors and the mean shape vectors. We assess the performance of our method using Weizmann human action dataset. Experimental results reveal that the proposed method performs generally very well on this dataset.

Variational Bayesian multinomial logistic Gaussian process classification

The multinomial logistic Gaussian process is a flexible non-parametric model for multi-class classification tasks. These tasks are often involved in solving a pattern recognition problem in real life. In such contexts, the multinomial logistic function (or softmax function) is usually assumed to be the likelihood function. But, exact inferences for this model have proved challenging problem because it requires high-dimensional integration. In this paper, we propose approximate variational Bayesian inference for the multinomial logistic Gaussian process model. First, we compute the second-order approximation for the logarithm of the logistic likelihood function using Taylor series expansion, and derive the posterior distributions of all hidden variables and model parameters using the variational Bayesian inference method. Second, we derive the predictive distribution of the latent classification variable corresponding to the relevant test data point using the characteristics of the Cauchy product for a standard Gaussian process using a learning model parameter. We conducted experiments to verify the effectiveness of the proposed model using a number of synthetic and real datasets. The results show that the proposed model has superior classification capability to existing methods.

Multiclass Data Classification Using Multinomial Logistic Gaussian Process Model

We propose the multiclass data classification method using Bayesian logistic Gaussian process model. First, we have defined the multinomial logistic Gaussian process classification model. Second, we have derived the predictive distribution of the classification variable corresponding to the new input data point by using a variational Bayesian inference method. Finally, in order to verify the performance of the proposed model, we conducted experiments using Iris real dataset. From the experimental results, we can see that the proposed model has achieved superior classification ability.

Human action recognition using variational Bayesian hidden Markov model with Gaussian-Wishart emission mixture model

In this paper, we proposed the human action recognition method using the variational Bayesian HMM with Gaussian — Wishart emission mixture model. First, we defined the Bayesian HMM based on a finite number of Gaussian-Wishart mixture components to support continuous emission observations. Second, we have considered a variational Bayesian inference method to derive the posterior distributions for hidden variables and parameters that are required to define the proposed model using training data. And then we have also derived the predictive distribution that is used to classify new action. Third, the human action classification using KTH data set has been conducted to evaluate the performance of proposed method. The experimental results showed that our method is more efficient with human action recognition than existing methods.

Human action classification using procrustes shape theory

In this paper, we propose new method that can classify a human action using Procrustes shape theory. First, we extract a pre-shape configuration vector of landmarks from each frame of an image sequence representing an arbitrary human action, and then we have derived the Procrustes fit vector for pre-shape configuration vector. Second, we extract a set of pre-shape vectors from tanning sample stored at database, and we compute a Procrustes mean shape vector for these preshape vectors. Third, we extract a sequence of the pre-shape vectors from input video, and we project this sequence of pre-shape vectors on the tangent space with respect to the pole taking as a sequence of mean shape vectors corresponding with a target video. And we calculate the Procrustes distance between two sequences of the projection pre-shape vectors on the tangent space and the mean shape vectors. Finally, we classify the input video into the human action class with minimum Procrustes distance. We assess a performance of the proposed method using one public dataset, namely Weizmann human action dataset. Experimental results reveal that the proposed method performs very good on this dataset.

A system architecture for real time traffic monitoring in foggy video

In foggy video, the visibility and contrast of objects are decreased dramatically, which causes the performance degradation of traffic monitoring systems. In this paper, an architecture for real-time traffic monitoring system is presented for foggy video. For the real-time traffic monitoring, it is required to satisfy two major constraints. First, the quality of an image after fog removal is good enough for further processing such as object detection and tracking. Second, it has to be computationally cheap for real-time processing. In this paper, a parallel architecture is proposed, consisting of N threads, for a real-time traffic monitoring system. The proposed parallel architecture shows the significant reduction of processing time for the development of real-time traffic monitoring systems.

Variational Bayesian Inference for Multinomial Dirichlet Gaussian Process Classification Model

In this paper, we propose the variational Bayesian inference algorithm which can drive approximate posterior distributions of both three latent functions and two parameters needed to define the multinomial Dirichlet Gaussian process (GP) classification model. This model consists of three components: a latent function with GP prior, a response function with multiclass, and a link function that relates the latent function and response mean. Here, we consider the variational Bayesian estimation method to estimate the proposed model. This is performed in two parts: one is to derive the variational posterior distribution of auxiliary variables and latent function, another is to derive the variational posterior distribution for the various parameters. Moreover, we have proposed a classification rule that can predict a particular category for a new observation by using the trained model. Finally, we conducted experiment using a well-known Iris data in order to verify the performance of the proposed model. Experimental result reveals that the proposed model shows good performance on this data set.

New Inference for a Multiclass Gaussian Process Classification Model using a Variational Bayesian EM Algorithm and Laplace Approximation

In this study, we propose a new inference algorithm for a multiclass Gaussian process classification model using a variational EM framework and the Laplace approximation (LA) technique. This is performed in two steps, called expectation and maximization. First, in the expectation step (E-step), using Bayes’ theorem and the LA technique, we derive the approximate posterior distribution of the latent function, indicating the possibility that each observation belongs to a certain class in the Gaussian process classification model. In the maximization step, we compute the maximum likelihood estimators for hyper-parameters of a covariance matrix necessary to define the prior distribution of the latent function by using the posterior distribution derived in the E-step. These steps iteratively repeat until a convergence condition is satisfied. Moreover, we conducted the experiments by using synthetic data and Iris data in order to verify the performance of the proposed algorithm. Experimental results reveal that the proposed algorithm shows good performance on these datasets.