Multivariate Adaptive Gaussian Mixture for Scene Level Anomaly Modeling

Video anomaly event detection is the process of finding an abnormal event deviation compared with the majority of normal or usual events. The main challenges are the high structure redundancy and the dynamic changes in the scenes that are in surveillance videos. To address these problems, we present a framework for anomaly detection and localization in videos that is based on locality-constrained affine subspace coding (LASC) and a model updating procedure. In our algorithm, LASC attempts to reconstruct the test sample by its top-k nearest subspaces, which are obtained by segmenting the normal samples space using a clustering method. A sample with a large reconstruction cost is detected as abnormal by setting a threshold. To adapt to the scene changes over time, a model updating strategy is proposed. We experiment on two public datasets: the UCSD dataset and the Avenue dataset. The results demonstrate that our method achieves competitive performance at a 700 fps on a single desktop PC.

As video data are occupying an increasingly more significant portion of global data traffic, video communication has become an indispensable component for most multimedia applications. As an enabling technology of video communication, although video coding is well standardized, the strategy to control video codec is highly customized to applications. The consistency of video quality is being paid more attention in many emerging applications. For example, in video surveillance, the quality of video frames should be stable in order to ensure the performance of machine intelligence algorithm, such as object detection precision. In this paper, we focus on achieving constant objective reconstruction quality in the process of video coding. To achieve certain rate-distortion performance, bit rate and distortion metric are usually modeled as a function of video content and control parameters of codec. For content modeling in existing work, there is still room for improvement, including the design of more efficient content feature, the compensation for assumption about constant RD characteristics among consecutive frames, and the adjustment of Lagrangian multiplier $\lambda $ according to content property. The main contributions of this paper are: 1) a robust content adaptive model for residual bit rate modeling based on content statistics called mean absolute partial transformed difference (MAPTD); 2) a content-related header bit rate modeling; 3) preprocessing scheme for robust content feature estimation at scene change; 4) a distortion model consistent with local content; and 5) content adaptive $\lambda $ determination. The experimental results show that our constant quality control strategy can achieve superior performance compared with the state-of-the-art algorithms.

These days consciousness about public safety increases and CCTV cameras are installed at almost all public places. But generally automatic smart surveillance systems are not available. In this manuscript, emphasis is given to detect and classify abnormal events in surveillance video especially in crowd environments. Abnormal event detection is a challenging task because the definition of abnormality is subjective. A normal event in one situation can be considered an abnormal event in another case. In the surveillance video with a dense crowd, automatic anomaly detection becomes very difficult because of clutter and severe occlusion. This manuscript represents CNN (Convolutional Neural Network) and CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) based approaches for detection and classification of abnormal events. The CNN architecture is developed from scratch and can be used for spatial domains. LSTM architecture is developed for the temporal domain. Feature sequences are generated using CNN model and given as input to LSTM model. Experiments are carried out using five different publicly available benchmark datasets. The performance is measured by accuracy and area under the ROC (receiver operating characteristic) curve (AUC). CNN-LSTM approach works better than only CNN.

The key for efficient dynamic spectrum access (DSA) is to model the spectral resources accurately. A large number of measurement campaigns have been performed to estimate the spectrum usage in outdoor and indoor scenarios. This spectrum usage estimation helps policy makers to optimize the spectrum management methodologies. The spectrum usage studies also assist researchers to constitute a way for efficient DSA using prior knowledge of the distribution of the observed data traffic in cognitive radio (CR) systems. In this paper we extend our previous work which statistically modeled the observed data traffic in the industrial, scientific and medical (ISM) band at 2.4 -GHz in two neighboring frequency subbands and time slots, respectively, to three neighboring frequency subbands and time slots, respectively. As before, the frequency and time correlation functions of the observed data traffic are modeled by an exponentially decaying function. The multivariate Gaussian mixture (MGM) is validated as a good candidate to model the joint distribution of measured data and also to estimate the correlation between the measured data in neighboring frequency subbands and as well as in time domain samples.

Copula based factorization in Bayesian multivariate infinite mixture models

Bayesian networks (BNs) are being studied in recent years for system diagnosis, reliability analysis, and design of complex engineered systems. In several practical applications, BNs need to be learned from available data before being used for design or other purposes. Current BN learning algorithms are mainly developed for networks with only discrete variables. Engineering design problems often consist of both discrete and continuous variables. This paper develops a framework to handle continuous variables in BN learning by integrating learning algorithms of discrete BNs with Gaussian mixture models (GMMs). We first make the topology learning more robust by optimizing the number of Gaussian components in the univariate GMMs currently available in the literature. Based on the BN topology learning, a new multivariate Gaussian mixture (MGM) strategy is developed to improve the accuracy of conditional probability learning in the BN. A method is proposed to address this difficulty of MGM modeling with data of mixed discrete and continuous variables by mapping the data for discrete variables into data for a standard normal variable. The proposed framework is capable of learning BNs without discretizing the continuous variables or making assumptions about their conditional probability densities (CPDs). The applications of the learned BN to uncertainty quantification and model calibration are also investigated. The results of a mathematical example and an engineering application example demonstrate the effectiveness of the proposed framework.

Maneuvering Target Tracking in Dense Clutter Based on Particle Filtering

We consider the problem of estimating a dense depth map from a single monocular image. Inspired by psychophysical evidence of visual processing in human vision systems (HVS) and natural scene statistics (NSS) models of image and range, we propose a Bayesian framework to recover detailed 3D scene structure by exploiting the statistical relationships between local image features and depth variations inherent in natural images. By observing that similar depth structures may exist in different types of luminance/chrominance textured regions in natural scenes, we build a dictionary of canonical range patterns as the prior, and fit a multivariate Gaussian mixture (MGM) model to associate local image features to different range patterns as the likelihood. Compared with the state-of-the-art depth estimation method, we achieve similar performance in terms of pixel-wise estimated range error, but superior capability of recovering relative distant relationships between different parts of the image.

In this paper, a robust and time-saving method for moving object detection in video sequences is proposed. Unlike methods based on complex background updating models which are computationally expensive, the proposed method can segment moving targets in real-time. It mainly includes three steps. In the first step, the background image is reconstructed by using the long-term and short-term background updating algorithms. The long-term updating algorithm detects the noisy motion regions and ghosts, while the short-term updating algorithm models the background pixel values with single Gaussian distributions, which can deal with slow lighting changes. In the second step, the shadows are removed via the color space based approach. Finally, different targets are located with the projection method. Experimental results prove that the presented method is robust to background noisy motions, shadows and scene changes, and can segment multiple objects precisely and quickly.

In this paper, we describe environment compensation approach based on MAP (maximum a posteriori) estimation assuming that the noise can be modeled as a single Gaussian distribution. It employs the prior information of the noise to deal with environmental variabilities. The acoustic-distorted environment model in the cepstral domain is approximated by the truncated first-order vector Taylor series(VTS) expansion and the clean speech is trained by using Self-Organizing Map (SOM) neural network with the assumption that the speech can be well represented as the multivariate diagonal Gaussian mixtures model (GMM). With the reasonable environment model approximation and effective clustering for the clean model, the noise is well refined using batch-EM algorithm under MAP criterion. Experiment with large vocabulary speaker-independent continuous speech recognition shows that this approach achieves considerable improvement on recognition performance.

This paper proposes a method for analysing natural outdoor scenes to find out changes due to slow landslide movements. The analysis is carried out by examining sequences of images and by matching the contours detected in each image, with a set of contours assumed as the model. Once matching is achieved, motion parameters are computed. If a movement is recognized, displacements are estimated in the image plane. The procedure applied to some test sequences has proved itself to be robust with respect to noise and spurious changes due to variable environmental conditions.

A major portion of the big data that is produced comprises of videos coming from surveillance cameras deployed to view streets, buildings, offices etc. The surveillance videos are mainly used for monitoring day to day activities. The video sequences are long and the events of interest occur only over a short duration. Hence, there is a pressing need to analyze and detect events to avoid continuous manual monitoring of entire video sequence. The first step towards that is to extract the foreground information. In this paper we present an effective online multilinear subspace learning algorithm which incrementally learns and models the background as a low-rank tensor. This background modeling combined with appropriate post processing steps is useful to detect anomalous events, thus in turn the foreground, in the video. The efficacy of the proposed method is also brought out in the simulation results provided.

Gaussian mixture model with the standard EM algorithm is a widely applicable approach for science computing and research. The truncation or censoring phenomenon is most frequently appearing in data collection work of science research. So this motivated interest in solving the truncated or censored data problem. This paper studied a truncated and censored EM algorithm for fltting multivariate Gaussian mixture models to data that is truncated and censored simultaneously. A split and merge operation was proposed to improve its sensitive to initialization. We verifled this method on some synthetic and real data sets and got some good result.

Guangdong Power Grid Corporation has put forward and implemented based on IP network digital, Linux, industrial grade layered architecture of substation video and environment surveillance system. In order to improve video coding efficiency and meet requirements of users on HD images resolution of substation video and environment surveillance system, this paper proposes an improved implicit motion information inter prediction algorithm for P slice in H.264 based on the spatio-temporal adaptive localized learning model. This algorithm essentially takes advantage of the temporal and spatial correlations that may exist within frames at the block level, and reduces the bits required for encoding motion information while retaining or even improving encoding quality. The simulation results show that the proposed algorithm improves encoding efficiency compared with the H.264 standard with 0.10 dB (QCIF) and 0.17 dB (WQVGA) average PSNR gain, with relatively increase in complexity.

Moving object detection is vitally used in video surveillance applications. Traditional Gaussian mixture model (GMM) based background subtraction (BGS) methods are usually performs well when background is stationary. However, they require parameter tuning to deal with dynamic backgrounds, whose background pixel values change over time. Particularly, the threshold which determines the pixels associated with moving objects from the resultant of BGS. To tackle this problem there is no ultimate solution. Considering that, this paper intents to present a novel idea to update the threshold of GMM based BGS with respect to color distortion, similarity and illumination measures in pixel level. Extensive experiments were carried out to demonstrate the effectiveness of the proposed method in comparison to some of the long-familiar GMM based BGS methods in literature. However, note that this paper is not attempted to provide a real-time technique, but rather to investigate the potential utilization of the aforementioned measures to set a threshold automatically to detect moving objects in video sequences.