Multi-domain anomaly detection in spatial datasets

A spatial anomaly captures a phenomenon occurring in a region which is vastly deviant in behavior with respect to the other normal observations. However, in reality this anomaly may impact other phenomena in the region across multiple domains, for example, crime is often linked to other sociopolitical factors or phenomenon such as poverty and education. Similarly, accidents in the region may be linked to other environmental factors such as weather and surface condition. So, finding anomalies across multiple domains is important in various applications. In this paper, we propose an approach for finding such a tangible anomalous window across multiple domains where window refers to the set of contiguous points in space, and since the window is multi-domain, there are several overlapping windows in the same space across domains. Our approach for finding anomalous window across the domains comprises the following steps: (1) single-domain anomaly detection: discovering anomalous window in each domain; (2) association rule mining: discovering relationship between the anomalous windows across domains using association rule mining; and (3) validation: validating the result using (a) Monte Carlo simulation, (b) correlation using lift and (c) ground truth evaluation. In addition, we also provide a probabilistic framework to evaluate the relationships between the spatial nodes as a postprocessing step. Finally, we provide a visualization technique for viewing the multi-domain anomalous window and the probabilistic relationships between the nodes. We provide detailed experimental results and comparisons with other approaches using real-world health ranking [51] and transportation datasets [50] with known ground truth windows. The results show that our approach is effective in finding the anomalies in multiple domains as compared to other approaches.     

Handling occlusions in video‐based augmented reality using depth information

Augmented Reality (AR) composes virtual objects with real scenes in a mixed environment where human–computer interaction has more semantic meanings. To seamlessly merge virtual objects with real scenes, correct occlusion handling is a significant challenge. We present an approach to separate occluded objects in multiple layers by utilizing depth, color, and neighborhood information. Scene depth is obtained by stereo cameras and two Gaussian local kernels are used to represent color, spatial smoothness. These three cues are intelligently fused in a probability framework, where the occlusion information can be safely estimated. We apply our method to handle occlusions in video‐based AR where virtual objects are simply overlapped on real scenes. Experiment results show the approach can correctly register virtual and real objects in different depth layers, and provide a spatial‐awareness interaction environment. Copyright © 2009 John Wiley & Sons, Ltd.     

SOLE: scalable on-line execution of continuous queries on spatio-temporal data streams

This paper presents the scalable on-line execution (SOLE) algorithm for continuous and on-line evaluation of concurrent continuous spatio-temporal queries over data streams. Incoming spatio-temporal data streams are processed in-memory against a set of outstanding continuous queries. The SOLE algorithm utilizes the scarce memory resource efficiently by keeping track of only the significant objects. In-memory stored objects are expired (i.e., dropped) from memory once they become insignificant. SOLE is a scalable algorithm where all the continuous outstanding queries share the same buffer pool. In addition, SOLE is presented as a spatio-temporal join between two input streams, a stream of spatio-temporal objects and a stream of spatio-temporal queries. To cope with intervals of high arrival rates of objects and/or queries, SOLE utilizes a load-shedding approach where some of the stored objects are dropped from memory. SOLE is implemented as a pipelined query operator that can be combined with traditional query operators in a query execution plan to support a wide variety of continuous queries. Performance experiments based on a real implementation of SOLE inside a prototype of a data stream management system show the scalability and efficiency of SOLE in highly dynamic environments. This work was supported in part by the National Science Foundation under Grants IIS-0093116, IIS-0209120, and 0010044-CCR.     

Spatio-temporal polygonal clustering with space and time as first-class citizens

Detecting spatio-temporal clusters, i.e. clusters of objects similar to each other occurring together across space and time, has important real-world applications such as climate change, drought analysis, detection of outbreak of epidemics (e.g. bird flu), bioterrorist attacks (e.g. anthrax release), and detection of increased military activity. Research in spatio-temporal clustering has focused on grouping individual objects with similar trajectories, detecting moving clusters, or discovering convoys of objects. However, most of these solutions are based on using a piece-meal approach where snapshot clusters are formed at each time stamp and then the series of snapshot clusters are analyzed to discover moving clusters. This approach has two fundamental limitations. First, it is point-based and is not readily applicable to polygonal datasets. Second, its static analysis approach at each time slice is susceptible to inaccurate tracking of dynamic cluster especially when clusters change over both time and space. In this paper we present a spatio-temporal polygonal clustering algorithm known as the Spatio-Temporal Polygonal Clustering (STPC) algorithm. STPC clusters spatial polygons taking into account their spatial and topological properties, treating time as a first-class citizen, and integrating density-based clustering with moving cluster analysis. Our experiments on the drought analysis application, flu spread analysis and crime cluster detection show the validity and robustness of our algorithm in an important geospatial application.     

Learning Behavioural Context

We propose a novel framework for automatic discovering and learning of behavioural context for video-based complex behaviour recognition and anomaly detection. Our work differs from most previous efforts on learning visual context in that our model learns multi-scale spatio-temporal rather than static context. Specifically three types of behavioural context are investigated: behaviour spatial context, behaviour correlation context, and behaviour temporal context. To that end, the proposed framework consists of an activity-based semantic scene segmentation model for learning behaviour spatial context, and a cascaded probabilistic topic model for learning both behaviour correlation context and behaviour temporal context at multiple scales. These behaviour context models are deployed for recognising non-exaggerated multi-object interactive and co-existence behaviours in public spaces. In particular, we develop a method for detecting subtle behavioural anomalies against the learned context. The effectiveness of the proposed approach is validated by extensive experiments carried out using data captured from complex and crowded outdoor scenes.     

FRaC: a feature-modeling approach for semi-supervised and unsupervised anomaly detection

Anomaly detection involves identifying rare data instances (anomalies) that come from a different class or distribution than the majority (which are simply called “normal” instances). Given a training set of only normal data, the semi-supervised anomaly detection task is to identify anomalies in the future. Good solutions to this task have applications in fraud and intrusion detection. The unsupervised anomaly detection task is different: Given unlabeled, mostly-normal data, identify the anomalies among them. Many real-world machine learning tasks, including many fraud and intrusion detection tasks, are unsupervised because it is impractical (or impossible) to verify all of the training data. We recently presented FRaC, a new approach for semi-supervised anomaly detection. FRaC is based on using normal instances to build an ensemble of feature models, and then identifying instances that disagree with those models as anomalous. In this paper, we investigate the behavior of FRaC experimentally and explain why FRaC is so successful. We also show that FRaC is a superior approach for the unsupervised as well as the semi-supervised anomaly detection task, compared to well-known state-of-the-art anomaly detection methods, LOF and one-class support vector machines, and to an existing feature-modeling approach.     

Managing sensor traffic data and forecasting unusual behaviour propagation

Sensor data on traffic events have prompted a wide range of research issues, related with the so-called ITS (Intelligent Transportation Systems). Data are delivered for both static (fixed) and mobile (embedded) sensors, generating large and complex spatio-temporal series. This scenario presents several research challenges, in spatio-temporal data management and data analysis. Management issues involve, for instance, data cleaning and data fusion to support queries at distinct spatial and temporal granularities. Analysis issues include the characterization of traffic behavior for given space and/or time windows, and detection of anomalous behavior (either due to sensor malfunction, or to traffic events). This paper contributes to the solution of some of these issues through a new kind of framework to manage static sensor data. Our work is based on combining research on analytical methods to process sensor data, and data management strategies to query these data. The first aspect is geared towards supporting pattern matching. This leads to a model to study and predict unusual traffic behavior along an urban road network. The second aspect deals with spatio-temporal database issues, taking into account information produced by the model. This allows distinct granularities and modalities of analysis of sensor data in space and time. This work was conducted within a project that uses real data, with tests conducted on 1,000 sensors, during 3 years, in a large French city.     

A Comparison of Inferences about Containers and Surfaces in Small-Scale and Large-Scale Spaces

Inference mechanisms about spatial relations constitute an important aspect of spatial reasoning as they allow users to derive unknown spatial information from a set of known spatial relations. When formalized in the form of algebras, spatial-relation inferences represent a mathematically sound definition of the behavior of spatial relations, which can be used to specify constraints in spatial query languages. Current spatial query languages utilize spatial concepts that are derived primarily from geometric principles, which do not necessarily match with the concepts people use when they reason and communicate about spatial relations. This paper presents an alternative approach to spatial reasoning by starting with a small set of spatial operators that are derived from concepts closely related to human cognition. This cognitive foundation comes from the behavior of image schemata, which are cognitive structures for organizing people's experiences and comprehension. From the operations and spatial relations of a small-scale space, a container–surface algebra is defined with nine basic spatial operators—inside, outside, on, off, their respective converse relations—contains, excludes, supports, separated_from, and the identity relation equal. The container–surface algebra was applied to spaces with objects of different sizes and its inferences were assessed through human-subject experiments. Discrepancies between the container–surface algebra and the human-subject testing appear for combinations of spatial relations that result in more than one possible inference depending on the relative size of objects. For configurations with small- and large-scale objects larger discrepancies were found because people use relations such as part of and at in lieu of in. Basic concepts such as containers and surfaces seem to be a promising approach to define and derive inferences among spatial relations that are close to human reasoning.     

基于目标时空上下文融合的视频异常检测算法

视频异常检测旨在发现视频中的异常事件,异常事件的主体多为人、车等目标,每个目标都具有丰富的时空上下文信息,而现有检测方法大多只关注时间上下文,较少考虑代表检测目标和周围目标之间关系的空间上下文。提出一种融合目标时空上下文的视频异常检测算法。采用特征金字塔网络提取视频帧中的目标以减少背景干扰,同时计算相邻两帧的光流图,通过时空双流网络分别对目标的RGB帧和光流图进行编码,得到目标的外观特征和运动特征。在此基础上,利用视频帧中的多个目标构建空间上下文,对目标外观和运动特征重新编码,并通过时空双流网络重构上述特征,以重构误差作为异常分数对外观异常和运动异常进行联合检测。实验结果表明,该算法在UCSD-ped2和Avenue数据集上帧级AUC分别达到98.5%和86.3%,在UCSD-ped2数据集上使用时空双流网络相对于只用时间流和空间流网络分别提升5.1和0.3个百分点,采用空间上下文编码后进一步提升1个百分点,验证了融合方法的有效性。     

Direction finding of moving ferromagnetic objects inside water by magnetic anomaly

The remote sensing methods by the use of magnetic anomaly are gaining importance in applications of defense technologies and industrial purposes recently. In this study, it is aimed to determine the remote detection, the variation of characteristic of the voltage in the sensor relative to the motion, the effects of material length, magnetic permeability and direction of motion of the object on this characteristic and to convert them to a useful mathematical expression by using magnetic anomaly of ferromagnetic objects such as submarines moving inside water. For this purpose, first of all, a water tank of 1 m³ is prepared and approximately a homogeneous magnetic field of 10⁻³ T is created within this water tank. Ferromagnetic materials with six different lengths and permeabilities are moved in three different directions relative to the position of the sensor by means of a computer controlled x–y scanner designed for this experiment inside this magnetic field. The magnetic change caused by this motion at the point where the sensor is positioned is detected as the output voltage of the sensor. A mathematical expression is formulated taken into account the variations of the sensor output voltage with respect to the length, magnetic permeability and the direction of motion of the material and it is validated by the experimental results. This study clearly shows that the existence and the direction of motions of ferromagnetic objects with different lengths and magnetic permeabilities inside water can be detected with high accuracy.     

Mining trajectories of moving dynamic spatio-temporal regions in sensor datasets

When mining large spatio-temporal datasets, interesting patterns typically emerge where the dataset is most dynamic. These dynamic regions can be characterized by a location or set of locations that exhibit different behaviors from their neighbors and the time periods where these differences are most pronounced. Examples include locally intense areas of precipitation, anomalous sea surface temperature (SST) readings, and locally high levels of water pollution, to name a few. The focus of this paper is to find and analyze the pattern of moving dynamic spatio-temporal regions in large sensor datasets. The approach presented in this paper uses a measure of local spatial autocorrelation over time to determine how pronounced the difference in measurements taken at a spatial location is with those taken at neighboring locations. Dynamic regions are analyzed both globally, in the form of spatial locations and time periods that have the largest difference in local spatial autocorrelation, and locally, in the form of dynamic spatial locations for a particular time period or dynamic time periods for a particular spatial node. Then, moving dynamic regions are identified by determining the spatio-temporal connectivity, extent, and trajectory for groups of locally dynamic spatial locations whose position has shifted from one time period to the next. The efficacy of the approach is demonstrated on two real-world spatio-temporal datasets (a) NEXRAD precipitation and (b) SST. Promising results were found in discovering highly dynamic regions in these datasets depicting several real environmental phenomenon which are validated as actual events of interest.     

自适应邻域相关性的背景建模

目的 背景建模在计算机视觉领域中是检测、跟踪、行为学习和识别的基础,被广泛地应用于视频监控的运动目标检测。混合高斯（MOG）和Codebook是其中具有代表性的方法,但它们假设像素点间信息是独立的,只保留了时域信息而忽略了空域信息,使得模型对背景的描述局限于时间上的连续性。针对上述问题,提出了一种自适应邻域相关性的背景建模方法（ANC）。方法 ANC在保留原始方法时域信息建模特性的同时,增加对邻域模型的复用,同时利用计算结果反馈自适应调整邻域区域,提高对前景值判断的准确性。首先利用原始基于像素点的背景建模方法进行候选前景检测,然后将候选前景检测结果为前景点的像素与邻域像素点模型进行对比,若邻域范围存在匹配则为背景点,若不存在则为前景点;最后引入像素置信度概念,自适应调整邻域范围的大小。结果 与MOG和Codebook相比,在changedetection标准数据库上,ANC在ROC（受试者工作特征曲线）和度量值等方面的平均精度和F-measure都提高了7%以上。结论 自适应邻域相关性的背景建模方法适用于复杂多模态背景,克服了基于像素点背景建模方法假设的局限性。与普通基于像素点的背景建模方法相比,具有更好的鲁棒性和抗噪性,对复杂背景具有更强的适应性。     

无线传感器网络中异常读数检测算法研究*

针对无线传感器网络的离群点检测算法由于没有充分考虑数据的时空关联性和网络的分布特性,导致检测精度低、通信量大和计算复杂度高等局限,提出了基于时空关联的分布计算与过滤的在线离群点检测算法。该算法在各传感器节点上利用传感器读数的时间关联性生成候选离群点,并利用空间关联性对候选离群点进行过滤得到局部离群点,最终将所有传感器节点上的局部离群点集中到sink节点上获得全局离群点。利用时空关联性提高了检测精度,利用分布计算与过滤减少了通信量和计算量,理论分析和实验结果均表明该算法优于现有算法。     

Detecting localized homogeneous anomalies over spatio-temporal data

The last decade has witnessed an unprecedented growth in availability of data having spatio-temporal characteristics. Given the scale and richness of such data, finding spatio-temporal patterns that demonstrate significantly different behavior from their neighbors could be of interest for various application scenarios such as—weather modeling, analyzing spread of disease outbreaks, monitoring traffic congestions, and so on. In this paper, we propose an automated approach of exploring and discovering such anomalous patterns irrespective of the underlying domain from which the data is recovered. Our approach differs significantly from traditional methods of spatial outlier detection, and employs two phases—(i) discovering homogeneous regions, and (ii) evaluating these regions as anomalies based on their statistical difference from a generalized neighborhood. We evaluate the quality of our approach and distinguish it from existing techniques via an extensive experimental evaluation.     

An efficient location encoding method for moving objects using hierarchical administrative district and road network

Due to the rapid development in mobile communication technologies, the usage of mobile devices such as cell phone or PDA has increased significantly. As different devices require different applications, various new services are being developed to satisfy the needs. One of the popular services under heavy demand is the location-based service (LBS) that exploits the spatial information of moving objects per temporal changes. In order to support LBS well, in this paper, we investigate how spatio-temporal information of moving objects can be efficiently stored and indexed. In particular, we propose a novel location encoding method based on hierarchical administrative district information. Our proposal is different from conventional approaches where moving objects are often expressed as geometric points in two-dimensional space, (x, y). Instead, in ours, moving objects are encoded as one-dimensional points by both administrative district as well as road information. Our method becomes especially useful for monitoring traffic situation or tracing location of moving objects through approximate spatial queries.     

Optical-flow based on an edge-avoidance procedure

This paper presents a differential optical flow method which accounts for two typical motion-estimation problems: (1) flow regularization within regions of uniform motion while (2) preserving sharp edges near motion discontinuities i.e., where motion is multimodal by nature. The method proposed is a modified version of the well known Lucas–Kanade (LK) algorithm. While many edge-preserving strategies try to minimize the effect of outliers by using a line process or a robust function, our method takes a novel approach to solve the problem. Based on documented assumptions, our method computes motion with a classical least-squares fit on a local neighborhood shifted away from where motion is likely to be multimodal. In this way, the inherent bias due to multiple motion around moving edges is avoided instead of being compensated. This edge-avoidance procedure is based on the non-parametric mean-shift algorithm which shifts the LK integration window away from local sharp edges. Our method also locally regularizes motion by performing a fusion of local motion estimates. The regularization is made with a covariance filter which minimizes the effect of uncertainties due in part to noise and/or lack of texture. Our method is compared with other edge-preserving methods on image sequences representing different challenges.     

Efficient Implementation Techniques for Topological Predicates on Complex Spatial Objects

Topological relationships like overlap, inside, meet, and disjoint uniquely characterize the relative position between objects in space. For a long time, they have been a focus of interdisciplinary research as in artificial intelligence, cognitive science, linguistics, robotics, and spatial reasoning. Especially as predicates, they support the design of suitable query languages for spatial data retrieval and analysis in spatial database systems and geographical information systems. While, to a large extent, conceptual aspects of topological predicates (like their definition and reasoning with them) as well as strategies for avoiding unnecessary or repetitive predicate executions (like predicate migration and spatial index structures) have been emphasized, the development of robust and efficient implementation techniques for them has been largely neglected. Especially the recent design of topological predicates for all combinations of complex spatial data types has resulted in a large increase of their numbers and stressed the importance of their efficient implementation. The goal of this article is to develop correct and efficient implementation techniques of topological predicates for all combinations of complex spatial data types including two-dimensional point, line, and region objects, as they have been specified by different authors and in different commercial and public domain software packages. Our solution consists of two phases. In the exploration phase, for a given scene of two spatial objects, all topological events like intersection and meeting situations are summarized in two precisely defined topological feature vectors (one for each argument object of a topological predicate) whose specifications are characteristic and unique for each combination of spatial data types. These vectors serve as input for the evaluation phase which analyzes the topological events and determines the Boolean result of a topological predicate (predicate verification) or the kind of topological predicate (predicate determination) by a formally defined method called nine-intersection matrix characterization. Besides this general evaluation method, the article presents an optimized method for predicate verification, called matrix thinning, and an optimized method for predicate determination, called minimum cost decision tree. The methods presented in this article are applicable to all known complete collections of mutually exclusive topological predicates that are formally based on the well known nine-intersection model.

A system for learning statistical motion patterns

Analysis of motion patterns is an effective approach for anomaly detection and behavior prediction. Current approaches for the analysis of motion patterns depend on known scenes, where objects move in predefined ways. It is highly desirable to automatically construct object motion patterns which reflect the knowledge of the scene. In this paper, we present a system for automatically learning motion patterns for anomaly detection and behavior prediction based on a proposed algorithm for robustly tracking multiple objects. In the tracking algorithm, foreground pixels are clustered using a fast accurate fuzzy k-means algorithm. Growing and prediction of the cluster centroids of foreground pixels ensure that each cluster centroid is associated with a moving object in the scene. In the algorithm for learning motion patterns, trajectories are clustered hierarchically using spatial and temporal information and then each motion pattern is represented with a chain of Gaussian distributions. Based on the learned statistical motion patterns, statistical methods are used to detect anomalies and predict behaviors. Our system is tested using image sequences acquired, respectively, from a crowded real traffic scene and a model traffic scene. Experimental results show the robustness of the tracking algorithm, the efficiency of the algorithm for learning motion patterns, and the encouraging performance of algorithms for anomaly detection and behavior prediction.     

基于融合时空数据的车辆加油行为多视图深度异常检测框架

车辆加油时空数据多源异构、关系复杂,现有成熟的异常检测方法难以对时空离散的加油活动数据进行分析,因此提出基于融合时空数据的车辆加油行为多视图深度异常检测框架。首先基于统一概念模型（UCM）对静态信息和动态活动数据进行关联融合管理,然后从空间视图、时间视图和语义视图角度对时空数据进行编码和转换,最后基于三种视图构建深度时空异常分析检测框架。车辆加油时空数据集上的实验结果表明,多种异常检测方法在融合时空数据上均可取得更低均方根误差（RMSE）,平均降低10.73%,所提方法比现有主流方法中结果最好的长短时记忆网络（LSTM）的RMSE降低19.36%。在信用卡欺诈公开数据集上的实验结果表明,所提方法较之逻辑回归模型,马修斯系数（MCC）提高了32.78%。以上实验验证了所提方法的有效性。     

Spatio-Temporal Data Handling with Constraints

Most spatial information systems are limited to a fixed dimension (generally 2) which is not extensible. On the other hand, the emerging paradigm of constraint databases allows the representation of data of arbitrary dimension, together with abstract query languages. The complexity of evaluating queries though might be costly if the dimension of the objects is really arbitrary. In this paper, we present a data model, based on linear constraints, dedicated to the representation and manipulation of multidimensional data. In order to preserve a low complexity for query evaluation, we restrict the orthographic dimension of an object O, defined as the dimension of the components ^O ₁ ,..., O _n such that O=O ₁×...× O _n. This allows to process queries independently on each component, therefore achieving a satisfying trade-off between design simplicity, expressive power of the query language and efficiency of query evaluation. We illustrate these concepts in the context of spatio-temporal databases where space and time are the natural components. This data model has been implemented in the DEDALE system and a spatio-temporal application, with orthographic dimension 2, is currently running, thus showing the practical relevance of the approach.