Ingrid Visentini

Fusing uncertain knowledge and evidence for maritime situational awareness via Markov Logic Networks

The concepts of event and anomaly are important building blocks for developing a situational picture of the observed environment. We here relate these concepts to the JDL fusion model and demonstrate the power of Markov Logic Networks (MLNs) for encoding uncertain knowledge and compute inferences according to observed evidence. MLNs combine the expressive power of first-order logic and the probabilistic uncertainty management of Markov networks. Within this framework, different types of knowledge (e.g. a priori, contextual) with associated uncertainty can be fused together for situation assessment by expressing unobservable complex events as a logical combination of simpler evidences. We also develop a mechanism to evaluate the level of completion of complex events and show how, along with event probability, it could provide additional useful information to the operator. Examples are demonstrated on two maritime scenarios of rules for event and anomaly detection.

Diversity-aware classifier ensemble selection via f-score

Display Omitted F-score measure used to rank the performance members of classifier ensembles.F-score based classifiers selection strategy proposed for online classification.Diversity-aware approach. Good balance between accuracy and generalization.Explicit handling of asymmetric samples distributions.Allows real-time video tracking via classification. The primary effect of using a reduced number of classifiers is a reduction in the computational requirements during learning and classification time. In addition to this obvious result, research shows that the fusion of all available classifiers is not a guarantee of best performance but good results on the average. The much researched issue of whether it is more convenient to fuse or to select has become even more of interest in recent years with the development of the Online Boosting theory, where a limited set of classifiers is continuously updated as new inputs are observed and classifications performed. The concept of online classification has recently received significant interest in the computer vision community. Classifiers can be trained on the visual features of a target, casting the tracking problem into a binary classification one: distinguishing the target from the background.Here we discuss how to optimize the performance of a classifier ensemble employed for target tracking in video sequences. In particular, we propose the F-score measure as a novel means to select the members of the ensemble in a dynamic fashion. For each frame, the ensemble is built as a subset of a larger pool of classifiers selecting its members according to their F-score. We observed an overall increase in classification accuracy and a general tendency in redundancy reduction among the members of an f-score optimized ensemble. We carried out our experiments both on benchmark binary datasets and standard video sequences.

Cascaded online boosting

In this paper, we propose a cascaded version of the online boosting algorithm to speed-up the execution time and guarantee real-time performance even when employing a large number of classifiers. This is the case for target tracking purposes in computer vision applications. We thus revise the online boosting framework by building on-the-fly a cascade of classifiers dynamically for each new frame. The procedure takes into account both the error and the computational requirements of the available features and populates the levels of the cascade accordingly to optimize the detection rate while retaining real-time performance. We demonstrate the effectiveness of our approach on standard datasets.

Fusing multiple video sensors for surveillance

Real-time detection, tracking, recognition, and activity understanding of moving objects from multiple sensors represent fundamental issues to be solved in order to develop surveillance systems that are able to autonomously monitor wide and complex environments. The algorithms that are needed span therefore from image processing to event detection and behaviour understanding, and each of them requires dedicated study and research. In this context, sensor fusion plays a pivotal role in managing the information and improving system performance. Here we present a novel fusion framework for combining the data coming from multiple and possibly heterogeneous sensors observing a surveillance area.

On-line boosted cascade for object detection

On-line boosting is a recent advancement in the field of machine learning that has opened a new spectrum of possibilities in many diverse fields. With respect to a static strong classifier, the on-line algorithm updates the ensemble using new incoming samples. This idea has been successfully exploited in tasks such as detection and tracking as a classification problem with good results. Our purpose is to provide an efficient and robust framework to build a cascade of on-line updated classifiers that, speeding up the application time, allows the employment of a higher number of features, thus achieving better detection performance.

Data Fusion in Modern Surveillance

The performances of the systems that fuse multiple data coming from different sources are deemed to benefit from the heterogeneity and the diversity of the information involved. The rationale behind this theory is the capability of one source to compensate the error of another, offering advantages such as increased accuracy and failure resilience. While in the past ambient security systems were focused on the extensive usage of arrays of single-type sensors, modern scalable automatic systems can be extended to combine multiple information coming from mixed-type sources. All this data and information can be exploited and fused to enhance situational awareness in modern surveillance systems. From biometrics to ambient security, from robotics to military applications, the blooming of multi-sensor and heterogeneous-based approaches confirms the increasing interest in the data fusion field. In this chapter we want to highlight the advantages of the fusion of information coming from multiple sources for video surveillance purposes. We are thus presenting a survey of existing methods to outline how the combination of heterogeneous data can lead to better situation awareness in a surveillance scenario. We also discuss a new paradigm that could be taken into consideration for the design of next generation surveillance systems.

Multi-sensor Multi-cue Fusion for Object Detection in Video Surveillance

We here present a multi-sensor data fusion architecture that takes into account the performance of video sensors in detecting moving targets for video surveillance purposes. Target detection and tracking is performed via classification by an ensemble of classifiers learned online using heterogeneous features for each target. A novel approach is then used to estimate the position of the target on the ground plane map by temporally fusing likelihood maps, then by approximating likelihoods analytically by a Gaussian function, and eventually projecting and fusing the likelihood functions. Experimental results are shown on real-world video sequences.

Dynamic Models for People Detection and Tracking

In this paper we propose a real-time algorithm for detecting and tracking moving objects in a video sequence. Based on the on-line boosting framework, our algorithm is able to detect an object as a member of a class, e.g. pedestrian, then a specific model for each instance of the class can be built on-line allowing at the same time robust tracking and recognition of the particular instance as it leaves and re-enters the scene. Promising experimental results have been performed on standard video sequences.

Diversity-Based Classifier Selection for Adaptive Object Tracking

In this work we propose a novel pairwise diversity measure, that recalls the Fisher linear discriminant, to construct a classifier ensemble for tracking a non-rigid object in a complex environment. A subset of constantly updated classifiers is selected exploiting their capability to distinguish the target from the background and, at the same time, promoting independent errors. This reduced ensemble is employed in the target search phase, speeding up the application of the system and maintaining the performance comparable to state of the art algorithms. Experiments have been conducted on a Pan-Tilt-Zoom camera video sequence to demonstrate the effectiveness of the proposed approach coping with pose variations of the target.

Sensor management: a new paradigm for automatic video surveillance

In this paper we discuss the new paradigm of Sensor Management that could be taken into consideration for the design of next generation surveillance systems The paradigm is meant to optimize the sensing capabilities of a system by taking into account the state of the environment being observed along with contextual information that can drive the choice of sensing modalities and platforms We thus provide a brief account on how the Sensor Management concepts developed within the data fusion community could be applied in the design of the next generation of surveillance systems.