تشخیص رفتار غیر عادی انسان با استفاده از دوربین های متعدد

In this paper a bottom-up approach for human behaviour understanding is presented, using a multi-camera system. The proposed methodology, given a training set of normal data only, classifies behaviour as normal or abnormal, using two different criteria of human behaviour abnormality (short-term behaviour and trajectory of a person). Within this system an one-class support vector machine decides short-term behaviour abnormality, while we propose a methodology that lets a continuous Hidden Markov Model function as an one-class classifier for trajectories. Furthermore, an approximation algorithm, referring to the Forward Backward procedure of the continuous Hidden Markov Model, is proposed to overcome numerical stability problems in the calculation of probability of emission for very long observations. It is also shown that multiple cameras through homography estimation provide more precise position of the person, leading to more robust system performance. Experiments in an indoor environment without uniform background demonstrate the good performance of the system.

Motion analysis in video and particularly human behaviour understanding has attracted many researchers [24], mainly because of its fundamental applications, which include video indexing, virtual reality, human–computer interaction and smart surveillance. Smart surveillance in itself is one of the most challenging problems in computer vision. Its goal is to automatically model and identify human behaviours, calling for human attention only when a suspicious behaviour is detected. With the increasing number of cameras in many public areas, the related research becomes more appealing and is offered more application possibilities. This work deals with the classification of behaviours as normal or abnormal. Based on the remark that abnormal behaviour is considered to be rather infrequent (and thus abnormal), we choose to model normal behaviour and define as abnormal any behaviour deviating from that normality model. Our methodology applies two classification criteria: (1) short-term behaviour; (2) trajectory. The short-term behaviour refers to the type of behaviour that can be localized in a spatio-temporal sense, i.e. is brief and within restricted space. Examples of such behaviours are walking, standing still, running, moving abruptly, etc. In the related literature the aforementioned classification criteria are mostly treated separately and, furthermore, few works concentrate on learning only normal behaviours. The methodology provided herein provides the discrimination of anomaly due to abnormal short-term motion, as happens in the case of abrupt motion, as well as anomaly due to long-term motion, as in the case of abnormal trajectory. Recently, several researchers have dealt with the problem of anomaly detection, which is the process of behaviour classification as normal or abnormal. A variety of methods, ranging from fully supervised [9] and [10] to semi-supervised [36] and unsupervised systems [21], [22] and [18], have been proposed in existing literature, which we further review in Section 2. It should be noted, however, that most of the existing approaches do not use multi-camera information, except for [38], where multiple video streams are combined via a coupled Hidden Markov Model. Our methodology contributes in current research in several ways: • The presented approach reflects two different criteria of labelling an observed behaviour as normal or abnormal, since the final abnormality decision depends on the output of two different classifiers with independent inputs: short-term behaviour information and trajectory information. • The behaviours are classified according to the target object's position on the ground plane, based on homography (see Section 4) which provides higher accuracy compared to pure image-based techniques. 1 • We introduce a continuous Hidden Markov Model (cHMM) as an one-class classifier, using the notion of length-normalized log-probability (see Section 6.1). • A novel algorithm implementing a Forward Backward procedure for the emission probability estimation in HMMs is proposed, handling numerical instability resulting from long sequences (see Section 6.2). The rest of the paper is organized as follows. In Section 2 recent literature is reviewed, hinting as to the problems the proposed method tackles. Section 3 provides an overview of the proposed architecture. In Section 4 we explain briefly how homography is used to obtain information on the position of target objects on the ground plane. In Section 5 short-term behaviours are defined in terms of a set of extracted features. Section 5.2 describes in detail the classification process which is based on short-term behaviours. In Section 6, on the other hand, trajectories’ classification is presented by elaborating how we have used a continuous Hidden Markov Model as an one-class classifier (Section 6.1). As an added value, Section 6.2 contains the description and foundation of a modified algorithm for the Forward Backward procedure of probability estimation tackling long sequences in contemporary computers. Finally, in Section 7 we provide the experimental results and Section 8 concludes this paper through a brief discussion on the lessons learned.

In this paper, we have presented a set of theoretical and practical tools for the domain of behaviour recognition, which have been integrated within a unified, automatic, bottom-up system based on the use of multiple cameras performing human behaviour recognition in an indoor environment, without a uniform background. The approach's innovation is fourfold: • We propose the application of two different criteria of human behaviour's abnormality used within a single methodology that needs only normal data for training. • We have proven that the application of multiple cameras can be fruitful, when it comes to determining abnormality based on the trajectory. • We have presented a methodology that lets a continuous Hidden Markov Model function as an one-class classifier, with very promising experimental results. • We have accomplished to offer an alternative to the Forward Backward algorithm for the recognition step of cHMMs in order to overcome arithmetic underflow in the case of very long observation sequences, without loss of precision. Our experimental results demonstrated the good performance of the system in the task of recognizing human behaviour's abnormality in a somewhat noisy environment, with different scenarios of action and participation of different actors. The experiments were implemented in offline and real-time conditions, with similar results, implying the robustness of the method. Furthermore, experiments with a single camera version of the system provide us the incentive to consider another, more robust method for the fusion of data in order to improve performance. The multiple camera methodology has, so far, been tested on scenarios with only one object inside the scene, without taking account any interactions between actors. It would be worthwhile to further investigate the effectiveness of our system using more features, such as the distance of the object from each camera, in order to improve the motion-based discriminatory performance of the system. However, other methodologies could also be tested in the place of the SVM classifier.