Unifying background models over complex audio using entropy

In this paper we extend an existing audio background modelling technique, leading to a more robust application to complex audio environments. The determination of background audio is used as an initial stage in the analysis of audio for surveillance and monitoring applications. Knowledge of the background serves to highlight unusual or infrequent sounds. An existing modelling approach uses an online, adaptive Gaussian Mixture model technique that uses multiple distributions to model variations in the background. The method used to determine the background distributions of the GMM leads to a failure mode of the existing technique when applied to complex audio. We propose a method incorporating further information, the proximity of distributions determined using entropy, to determine a more complete background model. The method was successful in more robustly modelling the background for complex audio scenes.

Generative models for automatic recognition of human daily activities from a single triaxial accelerometer

In this work, we compare two generative models including Gaussian Mixture Model (GMM) and Hidden Markov Model (HMM) with Support Vector Machine (SVM) classifier for the recognition of six human daily activity (i.e., standing, walking, running, jumping, falling, sitting-down) from a single waist-worn tri-axial accelerometer signals through 4-fold cross-validation and testing on a total of thirteen subjects, achieving an average recognition accuracy of 96.43% and 98.21% in the first experiment and 95.51% and 98.72% in the second, respectively. The results demonstrate that both HMM and GMM are not only able to learn but also capable of generalization while the former outperformed the latter in the recognition of daily activities from a single waist worn tri-axial accelerometer. In addition, these two generative models enable the assessment of human activities based on acceleration signals with varying lengths.

Internet traffic clustering with side information

Internet traffic classification is a critical and essential functionality for network management and security systems. Due to the limitations of traditional port-based and payload-based classification approaches, the past several years have seen extensive research on utilizing machine learning techniques to classify Internet traffic based on packet and flow level characteristics. For the purpose of learning from unlabeled traffic data, some classic clustering methods have been applied in previous studies but the reported accuracy results are unsatisfactory. In this paper, we propose a semi-supervised approach for accurate Internet traffic clustering, which is motivated by the observation of widely existing partial equivalence relationships among Internet traffic flows. In particular, we formulate the problem using a Gaussian Mixture Model (GMM) with set-based equivalence constraint and propose a constrained Expectation Maximization (EM) algorithm for clustering. Experiments with real-world packet traces show that the proposed approach can significantly improve the quality of resultant traffic clusters. © 2014 Elsevier Inc.

Using geometric shape variations to create an inverse model for a sheet metal process

The output of the sheet metal forming process is subject to much variation. This paper develops a method to measure shape variation in channel forming and relate this back to the corresponding process parameter levels of the manufacturing set-up to create an inverse model. The shape variation in the channels is measured using a modified form of the point distribution model (also known as the active shape model). This means that channels can be represented by a weighting vector of minimal linear dimension that contains all the shape variation information from the average formed channel.

The inverse models were created using classifiers that related the weighting vectors to the process parameter levels for the blank holder force (BHF), die radii (DR) and tool gap (TG) of the parameters. Several classifiers were tested: linear, quadratic Gaussian and artificial neural networks. The quadratic Gaussian classifiers were the most accurate and the most consistent type of classifier over all the parameters.

Integrated health monitoring for reinforced concrete beams: an experimental study

Civil infrastructures begin to deteriorate once they are built and used. Detecting the damages in a structure to maintain its safety is a topic that has received considerable attention in the literature in recent years. In vibration-based methods, the first few modes are used to assess the locations and the amount of damage. However, a small number of the global modes are not sufficient to reliably detect minor damage in the structure. Also, a common limitation of these techniques is that they require a high-fidelity model of the structure to start with, which is usually not available. Recently, guided waves (GW) have been found as an effective and efficient way to detect incipient damages due to its capacity of relatively long propagation range as well as its flexibility in selecting sensitive mode-frequency combinations. In this paper, an integrated structural health monitoring test scheme is developed to detect damages in reinforced concrete (RC) beams. Each beam is loaded at the middle span progressively to damage. During each loading step, acoustic emission (AE) method is used as a passive monitoring method to catch the AE signals caused by the crack opening and propagation. After each loading step, vibration tests and guided wave tests are conducted as a combined active monitoring measure. The modal parameters and wave propagation results are used to derive the damage information. Experimental results show that the integrated method is efficient to detect incipient damages in RC structures.

Damage identification of steel beams using local and global methods

Structural condition monitoring methods can be generally classified as local and global. While the global method needs only a small number of sensors to measure the low-frequency structural vibration properties, the acquired information is often not sufficiently sensitive to minor damages in a structure. Local methods, on the other hand, could be very sensitive to minor damages but their detection range is usually small. To overcome the drawbacks and take advantage of both methods, an integrated condition monitoring system has been recently developed for structural damage detection, which combines guided wave and structural vibration tests. This study aims at finding a viable damage identification method for steel structures by using this system. First, a spectral element modelling method is developed, which can simulate both wave propagation and structural vibration properties. Then the model is used in updating analysis to identify crack damage. Extensive numerical simulations and model updating works are conducted. The experimental and numerical results suggest that simply combining the objective functions cannot provide better structural damage identification. A two-stage damage identification scheme is more suitable for identifying damage in steel beams.

Gaussian vault geometry : integrated approach in design and fabrication of physical prototypes

This paper reports the second part of a study on the digital design and fabrication of scaled architectural prototypes. The first paper reported techniques in the realization of a double curved vault surface, the Gaussian Vault. The aims of the research here further extend this body of knowledge to a better understanding of constructible components. It addresses the problem of fabricating complex curved forms through the integration of the basic building elements, skin and structure, to achieve a scaled physical prototype. The focus of the experimentation is to investigate the process from which a digital surface form is conceived, to its preparation for fabrication and eventual construction in the fashion of a scaled model or workable prototype.

Human behavior recognition with generic exponential family duration modeling in the hidden semi-Markov model

The ability to learn and recognize human activities of daily living (ADLs) is important in building pervasive and smart environments. In this paper, we tackle this problem using the hidden semi-Markov model. We discuss the state-of-the-art duration modeling choices and then address a large class of exponential family distributions to model state durations. Inference and learning are efficiently addressed by providing a graphical representation for the model in terms of a dynamic Bayesian network (DBN). We investigate both discrete and continuous distributions from the exponential family (Poisson and Inverse Gaussian respectively) for the problem of learning and recognizing ADLs. A full comparison between the exponential family duration models and other existing models including the traditional multinomial and the new Coxian are also presented. Our work thus completes a thorough investigation into the aspect of duration modeling and its application to human activities recognition in a real-world smart home surveillance scenario.

Optimization of gaussian fuzzy membership functions and evaluation of the monotonicity property of fuzzy inference systems

In this paper, two issues relating to modeling of a monotonicity-preserving Fuzzy Inference System (FIS) are examined. The first is on designing or tuning of Gaussian Membership Functions (MFs) for a monotonic FIS. Designing Gaussian MFs for an FIS is difficult because of its spreading and curvature characteristics. In this study, the sufficient conditions are exploited, and the procedure of designing Gaussian MFs is formulated as a constrained optimization problem. The second issue is on the testing procedure for a monotonic FIS. As such, a testing procedure for a monotonic FIS model is proposed. Applicability of the proposed approach is demonstrated with a real world industrial application, i.e., Failure Mode and Effect Analysis. The results obtained are analysis and discussed. The outcomes show that the proposed approach is useful in designing a monotonicity-preserving FIS model.

Incremental learning of temporally-coherent gaussian mixture models

In this paper we address the problem of learning Gaussian Mixture Models (GMMs) incrementally. Unlike previous approaches which universally assume that new data comes in blocks representable by GMMs which are then merged with the current model estimate, our method works for the case when novel data points arrive oneby- one, while requiring little additional memory. We keep only two GMMs in the memory and no historical data. The current fit is updated with the assumption that the number of components is fixed, which is increased (or reduced) when enough evidence for a new component is seen. This is deduced from the change from the oldest fit of the same complexity, termed the Historical GMM, the concept of which is central to our method. The performance of the proposed method is demonstrated qualitatively and quantitatively on several synthetic data sets and video sequences of faces acquired in realistic imaging conditions

A pose-wise linear illumination manifold model for face recognition using video

The objective of this work is to recognize faces using video sequences both for training and novel input, in a realistic, unconstrained setup in which lighting, pose and user motion pattern have a wide variability and face images are of low resolution. There are three major areas of novelty: (i) illumination generalization is achieved by combining coarse histogram correction with fine illumination manifold-based normalization; (ii) pose robustness is achieved by decomposing each appearance manifold into semantic Gaussian pose clusters, comparing the corresponding clusters and fusing the results using an RBF network; (iii) a fully automatic recognition system based on the proposed method is described and extensively evaluated on 600 head motion video sequences with extreme illumination, pose and motion pattern variation. On this challenging data set our system consistently demonstrated a very high recognition rate (95% on average), significantly outperforming state-of-the-art methods from the literature.

Incremental learning of temporally-coherent Gaussian mixture models

In this paper we address the problem of learning Gaussian Mixture Models (GMMs) incrementally. Unlike previous approaches which universally assume that new data comes in blocks representable by GMMs which are then merged with the current model estimate, our method works for the case when novel data points arrive one- by-one, while requiring little additional memory. We keep only two GMMs in the memory and no historical data. The current fit is updated with the assumption that the number of components is fixed which is increased (or reduced) when enough evidence for a new component is seen. This is deducedfrom the change from the oldest fit of the same complexity, termed the Historical GMM, the concept of which is central to our method. The performance of the proposed method is demonstrated qualitatively and quantitatively on several synthetic data sets and video sequences of faces acquired in realistic imaging conditions.