H. Buxton and Shaogang Gong. Visual surveillance in a dynamic and uncertain world. Arti#cial Intelligence, 78:431#459, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....real image sequences is, in fact, a complex and challenging task. Another interesting characteristic of these scenarios is the fact that they are symbolic. Scenarios recognized by our system can therefore be combined at the symbolic level using temporal logic and other event reasoning [10] 22] [6] to recognize more sophisticated activities (e.g. a long combination of activities of several individual mobile objects or activities at different scenes) The effectiveness of our behavior analysis system is still limited by the quality of detection and tracking of mobile objects. Our ....

H. Btuxton and S. Gong, "Visual Surveillance in a Dynamic and Uncertain World," Artificial Intelligence, vol. 78, nos. 1-2, pp. 431- 459, 1995.

....often less than perfect. Most of the early work on activity representation comes from the field of Artificial Intelligence (AI) 1, 2] The formalisms that have been employed include HMMs, logic programming and stochastic Partially supported by a Grant from DARPA ONR N00014 2 1 0809. grammars [3, 4, 5, 6, 7, 8, 9, 10, 11]. Many uncertaintyreasoning models have been actively pursued in the AI and image understanding literature, including Belief networks [12] Dempster Shafer theory [13] and truth maintenance systems (TMS) 14, 15, 16, 17, 18, 19] Computer vision based activity analysis algorithms have been ....

H. Buxton and S. Gong, "Visual surveillance in a dynamic and uncertain world," AI, pp. 431--459, 1995.

.... of their actions and interactions (e.g. 1] 12] Large research projects devoted to video surveillance research have been conducted in the United States (e.g. DARPA s Video Surveillance and Monitoring (VSAM) project [13] Europe (the ESPRIT PASSWORDS [14] AVS PV [15] and VIEWS [16] [17] projects) and Japan (the Cooperative Distributed Vision project [18] Automated surveillance has also been the topic of recent international workshops [19] 23] and special sections in journals [24] 25] In addition to the obvious security and traffic monitoring applications, other diverse ....

....event detection program that scans the log files for common events that can be given a semantic label. There has been much work on parsing sequences of low level surveillance observations, particularly time stamped trajectories, into events that signify object interactions [5] 10] 12] 15] [17], 87] 90] Our prototype event detector is based on hidden Markov models (HMMs) 10] 87] trained to recognize simple object interactions. Briefly, output from the low level detection and tracking surveillance algorithms is quantized into the following discrete set of attributes and values ....

H. Buxton and S. G. Gong, "Visual surveillance in a dynamic and uncertain world," Artif. Intell., vol. 78, pp. 431--459, Oct. 1995.

....of objects simultaneously with only elementary means. Like other Bayesian multiobject trackers [6,7,8,9,10,11] the presented system uses sequential Monte Carlo methods [12] for Bayesian inference. Though the idea of hierarchically organized prior knowledge can also be found in previous work, e.g. [13,14], the idea of a more general and versatile vision system is only marginal in these approaches. Tracking multiple objects simultaneously within a single tracker allows to model not only the behavior of each individual object but also their interactions. This additional capability can be exploited ....

Buxton, H., Gong, S.: Visual surveillance in a dynamic and uncertain world. Artificial Intelligence 78 (1995) 431--459

....in [6] where the feature extraction is distributed in a hypercube multicomputer network. Secondly, features can be extracted selectively, on a per need basis, in order to minimize feature extraction cost while maximizing the classification certainty (hypothesis driven feature extraction) In [7] a framework for visual surveillance is proposed. This framework selectively extracts features in order to focus the processing resources on relevant features. Likewise, hypothesis driven feature extraction is used in [3] to reduce the number of information sources required for reliable ....

H. Buxton and S. Gong, "Visual Surveillance in a Dynamic and Uncertain World," Artificial Intelligence, vol. 78, no. 1-2, pp. 431--459, 1995.

....being adherence to the causal rules applied to the system. There has been little work on applying uncertainty measures to high level vision tasks, particularly the analysis and interpretation of dynamic object interaction in image sequences [6] Recently, Huang et al. 19] and Buxton and Gong [11] used a Bayesian belief network and inference engine [2] in sequences of highway traffic scenes to produce high level concepts like car changing lane and car stalled. In general, belief networks propagate values around the network as vectors between events [28] Belief networks are regarded as ....

H. Buxton and S. G. Gong, "Visual surveillance in a dynamic and uncertain world," Artif. Intell., vol. 78, pp. 431--459, 1995.

....that there are at least two distinct difficulties involved in detecting relevant events: 1. The relevant events have to occur in sensor range. 2. Once in sensor range, the events have to be detected and recognised as relevant. Although the issue of automatic classification of events (see e.g. [1]) is extremely important and essential for artificial intelligence, we concentrate our research on the first difficulty. That is we are interested in strategies for moving the available sensors in such a manner that the relevant events are detected. In order to evaluate such surveillance ....

Hilary Buxton and Shaogang Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

....provide images of the humans in specific areas. 1 Introduction Automated monitoring of human activity is important for many applications. The problem of analyzing human activity in video has been the focus of several researchers effort and several systems have been described in the literature [1, 2, 3, 4, 5, 6, 7, 8, 9]. Many of these systems consist of a computer vision system to detect and segment a moving object and a higher level interpretation module. In very specialized applications other sensors are used besides vision. Automatic interpretation of the data is very difficult and most systems in use ....

H. Buxton and S.G. Gong. Visual surveillance in a dynamic and uncertain world. AI, 78(1-2):431--459, October 1995.

....with respect to the task. Generative graphical models such as the Bayesian Belief Network (BBN) or Hidden Markov Model (HMM) are widely used at a more cognitive level in visual processing since they support not only learning but also some kinds of contextual processing and task control, e.g. [6]. For an introduction to probabilistic reasoning in these models see [26] and for more variational learning methods see [21] Here we focus on the HMM for gesture analysis, which can be made sensitive to the detailed task context. One advantage of HMMs is that the hidden purposes of regular ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

....networks are an effective vehicle for combining user supplied semantics with conflicting and noisy observations to deduce an overall consistent interpretation of the scene. BBNs have been used previously as a framework for tracking multiple vehicles under occlusion using contextual information [4]. In [18] a naive BBN was used to characterise and classify objects in a visual scene. For tracking body parts under discontinuous motion the BBN framework is ideal because unlike other tracking methods such as Kalman filtering or CONDENSATION [12] that explicitly model the dynamics through ....

Hilary Buxton and Shaogang Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431459, 1995.

....Bayesian belief nets; Condensation; Discontinuous motion trajectories; Dynamic scene models; Pixel energy history; Segmentation; Semantics of visual behaviour. 1 Problem Statement Understanding visual behaviour is essential for the interpretation of human actions captured in image sequences [7,43,47]. Visual behaviours are often represented as structured patterns of visual events, e.g. ordered sequences or continuous trajectories of measurable imagery properties including object shape, colour and position. However, the key to understanding behaviour is to Corresponding author: sggdc s. qmw. ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431-459, 1995.

....known causal dependencies with estimated statistical knowledge. They are essentially providing closed loop control using both topdown and bottom up messages in the propagation of belief values. They also provide the possibility of learning and re ning visual representations by observation [15], 16] Bayes nets have been used in many demanding applications such as BATmobile [17] and TEA system [13] HMMs are also widely used in visual processing, as seen in the review of recent work on behaviour analysis below. The advantage here is that the hidden purposes of regular behaviour ....

....However, it can be argued that for application speci c tracking, there should be modeling of dependencies and explicit association of entities over time [26] More recently, Black and Fleet [27] have developed a full generative Bayesian framework for tracking motion boundaries. Buxton and Gong [15] have developed a systematic methodology for the design and integration of advanced vision systems using Bayes nets. These networks allow dynamic updating of values in evidence and interpretation nodes, but not speci cation of the temporal constraints themselves. Howarth and Buxton, as discussed ....

[Article contains additional citation context not shown here]

H. Buxton and S. Gong, \Visual surveillance in a dynamic and uncertain world," Arti cial Intelligence, vol. 78, pp. 431-459, 1995.

....with respect to the task. Generative graphical models such as the Bayesian Belief Network (BBN) or Hidden Markov Model (HMM) are widely used at a more cognitive level in visual processing since they support not only learning but also some kinds of contextual processing and task control, e.g. [5]. For an introduction to probabilistic reasoning in these models see [25] and for more variational learning methods see [20] Here we focus on the HMM for gesture analysis, which can be made sensitive to the detailed task context. One advantage of HMMs is that the hidden purposes of regular ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Art. Intelligence, 78:431--459, 1995.

....but computationally efficient techniques. Thus, our models are able to support partial view invariance, and are sufficient to recognise people s gestures in dynamic scenes. Such taskspecific representations need to be used to avoid unnecessary computational cost in dynamic scene interpretation [1, 19]. For our purposes, human behaviour can be considered to be any temporal sequence of body movements or configurations, such as a change in head pose, walking or waving. However, the human body is a complex, non rigid articulated system capable of almost infinite spatial and dynamic variations. ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431-459, 1995.

....approximate and computationally efficient RBF techniques, which support partial view invariance, sufficient to recognise people s expressions and gestures in dynamic scenes. Such task specific representations need to be used to avoid unnecessary computational cost in dynamic scene interpretation [5]. Most existing recurrent network models take a long time to train, but simple timedelay RBF networks provide a fast and effective method of identifying arbitrary behaviours [14] The main problem with this alternative strategy for learning behavioural models is that it is difficult to classify ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

....networks are an e ective vehicle for combining user supplied semantics with con icting and noisy observations to deduce an overall consistent interpretation of the scene. BBNs have been used previously as a framework for tracking multiple vehicles under occlusion using contextual information [4]. In [18] a naive BBN was used to characterise and classify objects in a visual scene. For tracking body parts under discontinuous motion the BBN framework is ideal because unlike other tracking methods such as Kalman ltering or CONDENSATION [12] that explicitly model the dynamics through ....

Hilary Buxton and Shaogang Gong. Visual surveillance in a dynamic and uncertain world. Articial Intelligence, 78:431-459, 1995.

....approximate and computationally efficient RBF techniques, which support partial view invariance, sufficient to recognise people s expressions and gestures in dynamic scenes. Such task specific representations need to be used to avoid unnecessary computational cost in dynamic scene interpretation [5]. Most existing recurrent network models take a long time to train, but simple timedelay RBF networks provide a fast and effective method of identifying arbitrary behaviours [15] The main problem with this alternative strategy for learning behavioural models is that it is difficult to classify ....

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

No context found.

H. Buxton and Shaogang Gong. Visual surveillance in a dynamic and uncertain world. Arti#cial Intelligence, 78:431#459, 1995.

No context found.

H. Buxton and S. Gong. Visual surveillance in dynamic and uncertain world. Artificial Intelligence Journal, 78:431 -- 459, 1995.

No context found.

H. Buxton and S. Gong, `Visual surveillance in dynamic and uncertain world', Artificial Intelligence Journal, 78, 431 -- 459, (1995).

No context found.

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, pages 431--459, 1995.

No context found.

Buxton, H., Gong, S.: Visual surveillance in a dynamic and uncertain world. Artificial Intelligence 78 (1995) 431--459

No context found.

H. Buxton and Shaogang Gong, "Visual Surveillance in a Dynamic and Uncertain World," Artificial Intelligence 78(1-2), pp. 431--459, 1995.

No context found.

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

No context found.

H. Buxton and S. Gong. Visual surveillance in a dynamic and uncertain world. Artificial Intelligence, 78:431--459, 1995.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC