Buxton, H., Gong, S.: Advanced Visual Surveillance using Bayesian Networks. In: International Conference on Computer Vision, Cambridge, Massachusetts (1995)  Home/Search   Document Details and Download   Summary   Related Articles   Check

....studies that describe behavior from image sequences. More recently, Bayesian belief networks surface as a viable method for detecting and describing interactions. The framework proposed by Buxton et al. uses Bayesian Networks to perform surveillance and evaluate evidence in well understood scenes [35]. In their approach, static and dynamic Bayesian networks provide robust tracking and segmentation. Using rules about the scene and human activities as constraints, these networks can be use to analyze user behaviors. Bayesian inference methods are also useful for making conclusions based on the ....

H. Buxton and S. Gong, \Advanced Visual Surveillance using Bayesian Networks," International Conference on Computer Vision, Cambridge, MA, June 1995.

.... for Carnegie Mellon s CyberScout semi autonomous, mobile video surveillance platform [15] 2 The Design Philosophy For many of the real time video surveillance systems discussed in the literature, the common objectives are to detect, classify and track objects of interest in the environment [1, 2, 6, 7, 8, 10]. Typically the video stream is analyzed by a series of processes that perform each of these tasks. The motion detector nominates candidate motion regions in each video frame. The tracker associates motion regions across video frames, producing image sequences for each candidate moving object. The ....

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using Bayesian networks. In Proceedings, IEEE Workshop on Context-Based Vision, 1995.

....an excess of parameters, and an associated overfitting of data when they are applied to reason about long and complex temporal sequences with limited training data. Finally, in recent years, more complex Bayesian networks have also been adopted for the modeling and recognition of human activities [15, 9, 6, 4, 11, 7]. To date, however, there has been little research on realtime, multimodal systems for HCI that use probabilistic methods to model typical human activities in a hierarchical manner. The methods and working system described in this paper focus on this representation. We show how with our approach ....

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, pages 111--123, Cambridge, Massachusetts, June 1995.

....outputs to the task variables. These probabilities are learned automatically from training data. While Bayesian network models are not yet in widespread use within the computer vision community, there is a growing body of work on their application to object recognition [11] scene surveillance [2], video analysis [22, 7] and selective perception [19] Much of this earlier work relies upon expert knowledge to instantiate network parameters. In contrast, we have explored the ability to learn network parameters from training data. Learning is a key step in fusing sensor outputs at the data ....

H. Buxton and S. Gong. Advanced visual surveillance using bayesian networks. In ICCV '95 Workshop on Context-Based Vision, pages 111--122, Cambridge MA, 1995.

.... plans [7] Further, they have been used to integrate action patterns and beliefs about an agent s mental state [38] Previous work in traffic understanding has used an agent based belief network and agentcentered features for recognition of driving activity from simulated [11] and real data [6, 19]. Unlike that work our task requires that the system must also represent the logical and temporal relationships between multiple agents. Remagnino, Tan, and Baker [41] described a pedestrian and car tracking and surveillance system that models the interaction between any two agents using a small ....

....knowledge engineer to modularize concepts. Goal detection networks are designed to primarily consider evidence local in space and time. 5. Deictic, or agent centered, goal detectors can manage the complexity of multiagent feature selection. This assumption has been used successfully in prior work [1, 6, 11]. We use features such as the closest agent instead of agent 5, so that some detectors can be used without an explicit search that matches data to objects. The remainder of this section describes the details of how these representational assumptions are used to recognize multiagent action. 6.2. ....

H. Buxton and S. Gong, Advanced visual surveillance using Bayesian networks, in Proc. of the Workshop on Context-Based Vision, Cambridge, MA, June 1995, pp. 111--123, IEEE Computer Society Press, LOS Alamitos, CA.

....concerned with detecting when interactions between people occur, and classifying the type of interaction. Over the last decade there has been growing interest within the computer vision and machine learning communities in the problem of analyzing human behavior in video ( 7] 1] 15] [5], 13] 9] 6] 8] Such systems typically consist of a low or mid level computer vision system to detect and segment a human or car, and a higher level interpretation module that classifies the motion into atomic behaviors such as, for example, a pointing gesture or a car turning left. ....

H. Buxton and S. Gong. Advanced visual surveillance using bayesian networks. In International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

.... plans [7] Further, they have been used to integrate action patterns and beliefs about an agent s mental state [38] Previous work in traffic understanding has used an agent based belief network and agentcentered features for recognition of driving activity from simulated [11] and real data [6, 19]. Unlike that work our task requires that the system must also represent the logical and temporal relationships between multiple agents. Remagnino, Tan, and Baker [41] described a pedestrian and car tracking and surveillance system that models the interaction between any two agents using a small ....

....engineer to modularize concepts. Goal detection networks are designed to primarily consider evidence local in space and time. 5. Deictic, or agent centered, goal detectors can manage the complexity of multi agent feature selection. This assumption has been used successfully in prior work [1, 11, 6]. We use features such as the closest agent instead of agent 5, so that some detectors can be used without an explicit search that matches data to objects. The remainder of this section describes the details of how these representational assumptions are used to recognize multi agent action. ....

H. Buxton and S. Gong. Advanced visual surveillance using Bayesian networks. In Proc. of the Workshop on Context-Based Vision, pages 111--123, Cambridge, MA, June 1995. IEEE Computer Society Press.

....outputs to the task variables. These probabilities are learned automatically from training data. While Bayesian network models are not yet in widespread use within the computer vision community, there is a growing body of work on their application to object recognition [11] scene surveillance [2], video analysis [22, 7] and selective perception [19] Much of this earlier work relies upon expert knowledge to instantiate network parameters. In contrast, we have explored the ability to learn network parameters from training data. Learning is a key step in fusing sensor outputs at the data ....

H. Buxton and S. Gong. Advanced visual surveillance using bayesian networks. In ICCV '95 Workshop on Context-Based Vision, pages 111--122, Cambridge MA, 1995.

....sequences of maneuvers and the goals to be achieved by such activities. Related publications by other groups have been surveyed in [ Nagel and Kollnig, 1995 ] as well as in [ Nagel, 1987 ] and [ Kollnig et al. 1994 ] A very useful recent discussion of this entire problem area can be found in [ Buxton and Gong, 1995 ] and [ Srihari, 1995 ] Space limitations cause us to refer the reader to these survey papers. Based on considerations discussed in [ Nagel, 1987 ] admissible sequences of vehicle maneuvers on the premise of a gas station have been represented by a formal language see Figure 1 [ Nagel, ....

H. Buxton, S. Gong, Advanced Visual Surveillance Using Bayesian Networks, in Proc. Workshop on Context-Based Vision, 19 June 1995, Cambridge/MA, pp. 111--122.

....as the primary means of recognizing objects. Bayesian methods provide a formal means to reason about partial beliefs under conditions of uncertainty [10] The framework proposed by Buxton et al. uses Bayesian Networks to perform surveillance and evaluate evidence in well understood scenes [3]. Yi and Chelberg assert the appropriateness of these networks for selecting probable objects based on discriminating features and domain specific knowledge [14] We incorporate similar concepts in our framework Figure 1. Structure for a book article. for summarizing activity and labeling unknown ....

H. Buxton and S. Gong, "Advanced Visual Surveillance using Bayesian Networks," International Conference on Computer Vision, Cambridge, Mass., June 1995.

.... time slice, there is only a single state variable and an observation node, is the wellknown Hidden Markov Model (HMM) 27] These Bayesian models have been applied by a number of researchers to the general problem of recognising and classifying individual or group behaviours in the spatial domain [5, 15, 11, 24, 3]. However, in all these applications, the domain is usually locally restricted, i.e. the behaviour of interest is assumed to take place within a single room, or within a local spatial region. Thus the need for dealing with di erent levels of abstraction does not arise. Modelling the agent s ....

....is in contrast to the Coupled Fatorial HMM where the nodes and links usually do not have a clear semantic causal interpretation. There have been a number of applications of Bayesian networks in dealing with noisy data in spatio temporal domain, including monitoring and surveillance of trac scenes [5, 15, 11], 5 tracking human movement and group behaviours [24] recognising and classifying human gestures [3] However, we note that in all these applications, the domains are usually locally restricted, e.g. the domain is restricted to a single room or a single ground space region. Thus, the need for ....

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using Bayesian networks. In Proceedings of the IEEE International Conference on Computer Vision, 1995.

....extraction, object recognition, corresponding problem, traffic analysis, route tracing I. INTRODUCTION Due to upcoming low cost hardware and the progress in algorithmic research, Computer Vision has become a promising base technology for traffic sensoring systems. Various research projects [1,5] pointed out advantages of video based technology over conventional systems. Among these advantages are portability, ease of installation, reliability, long lifetime and low cost. These advantages are standard features in integrated video based traffic analysis systems [3,6] Another major new ....

....manner, the lower and vertical boundaries are determined. V. FEATURE EXTRACTION AND RECOGNITION In order to realize the recognition, a function rec is defined as a measure for the similarity of two plate image feature vectors with the signature: rec: feature vector x feature vector [0,1] This function provides a unique value for the recognition of a vehicle at two survey points M and M j . It can be used as a weighting factor in the calculation of route related traffic data. Furthermore, every survey point M is assigned a unique identification number i. For every M i the set D ....

Buxton, H.; Gong, S.: Advanced Visual Surveillance using Bayesian Networks. In: International Conference on Computer Vision, Cambridge, Massachusetts, (1995).

....particularly concerned with detecting when interactions between people occur, and classifying the type of interaction. Over the last decade there has been growing interest within the computer vision and machine learning communities in the problem of analyzing human behavior in video ( 2] 3] 4] [5], 6] 7] 8] 9] Such systems typically consist of a low or mid level computer vision system to detect and segment a moving object human or car, for example , and a higher level interpretation module that classifies the motion into atomic behaviors such as, for example, a pointing ....

Hilary Buxton and Shaogang Gong, "Advanced visual surveillance using bayesian networks," in International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

.... using statistical Bayesian approaches [11] Bobick also presents several approaches to the machine perception of motion and discusses the role and levels of knowledge in each [1] The framework proposed by Buxton et al. uses Bayesian Networks to perform surveillance in well understood scenes [6]. Our approach attempts to extend much of this work by characterizing the relationship between human motion and environmental objects. 3 Methodology Our goal is to describe people s interactions with objects in as much detail as possible. We develop parameterized, dynamic classes for objects ....

H. Buxton and S. Gong, "Advanced Visual Surveillance using Bayesian Networks," International Conference on Computer Vision, Cambridge, Mass., June 1995.

....explicitly represent knowledge dependencies and are computationally well understood, for multi agent action recognition. Previous work in traffic understanding has used an agent based belief network and agent centered features for recognition of driving activity from simulated data[8] and real data[5, 13]. Unlike this work and other systems that reason about action using physics based properties[21] here the system must also represent and recognize high level domain knowledge requiring that logical and temporal relationships between multiple agents be modeled. Remagnino, Tan, and Baker recently ....

H. Buxton and S. Gong. Advanced visual surveillance using Bayesian networks. In Proc. of the Workshop on Context-Based Vision, 1995.

....et al. 66] presents a system designed to assess interactions between people using statistical Bayesian approaches. Bobick [11] also presents several approaches to the machine perception of motion and discusses the role and levels of knowledge in each. The framework proposed by Buxton et al. [22] uses Bayesian Networks to perform surveillance in well understood scenes. To test our representations, we are undertaking a few experiments in natural environments where people interact with their surroundings. In our first experiment, we used a real office environment equipped with typical ....

H. Buxton and S. Gong. Advanced visual surveillance using bayesian networks. In Proceedings of International Conference on Computer Vision, 1995.

.... Buxton 92] the VIEWS system is discussed which establishes in fact the feedback from the high level knowledge down to the image processing task in order to treat occlusions between temporarily occluded vehicles. While VIEWS focused on the handling of occlusion, Buxton Gong 95b] see, too, Buxton Gong 95a] present a broader approach for using high level knowledge in different domains within a visual surveillance system. These authors exploit several models in order to deal with uncertain and incomplete image information: a camera model, a (hierarchical) ground plane model, a (volumetric) object ....

H. Buxton and S. Gong, Advanced Visual Surveillance Using Bayesian Networks, in Proc. Workshop on Context-Based Vision, 19 June 1995, Cambridge/MA, pp. 111--122.

....concerned with detecting when interactions between people occur, and classifying the type of interaction. Over the last decade there has been growing interest within the computer vision and machine learning communities in the problem of analyzing human behavior in video ( 10] 3] 20] [8], 17] 14] 9] 11] Such systems typically consist of a low or mid level computer vision system to detect and segment a moving object human or car, for example , and a higher level interpretation module that classifies the motion into atomic behaviors such as, for example, a ....

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using bayesian networks. In International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

....concerned with detecting when interactions between people occur, and classifying the type of interaction. Over the last decade there has been growing interest within the computer vision and machine learning communities in the problem of analyzing human behavior in video ( 10] 3] 21] [8], 17] 14] 9] 11] Such systems typically consist of a low or mid level computer vision system to detect and segment a moving object human or car, for example , and a higher level interpretation module that classifies the motion into atomic behaviors such as, for example, a ....

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using bayesian networks. In International Conference on Computer Vision, Cambridge, Massachusetts, June 1995.

....vision applications is discussed. Keywords: RBF Networks, Time Delay Networks, Vision, Temporal Behaviours, Face Recognition, Image Sequences, View Invariance. 1 Introduction Recognising simple behaviours is an important capability for many computer vision applications, e.g. visual surveillance (Gong Buxton 1995) or biomedical sequence understanding (Psarrou Buxton 1993) The behaviour in the experiments reported in this paper is simply head rotation to the left or right. However, the work raises important issues for connectionist techniques: 1) time, 2) representation, and 3) learning with ....

Gong, S. & Buxton, H. (1995), Advanced visual surveillance using bayesian nets, in `IEEE Workshop on Context-Based Vision', Cambridge, MA.

....and face tracking from image sequences and achieves a high success rate when tested on sequences of known and unknown individuals with large viewpoint di#erences. 1. Introduction Trajectory prediction is an important capability for many computer vision applications, e.g. visual surveillance [3] or biomedical sequence understanding [11] Multi layer perceptrons with supervised learning are very popular for applications which can use static representations, but time is important in many domains, e.g. vision, speech and motor control. Dynamic neural networks can be constructed by adding ....

S. Gong and H. Buxton. "Advanced visual surveillance using Bayesian nets". In IEEE Workshop on Context-Based Vision, Cambridge, MA., 1995.

No context found.

Buxton, H., Gong, S.: Advanced Visual Surveillance using Bayesian Networks. In: International Conference on Computer Vision, Cambridge, Massachusetts (1995)

No context found.

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, 1995.

No context found.

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using Bayesian networks. In Proc. of IEEE Workshop on Context-Based Vision, 1995.

No context found.

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, 1995.

No context found.

H. Buxton and S. Gong, `Advanced visual surveillance using bayesian networks', in Workshop on Context-based Vision, Cambridge, (1995). IEEE.

No context found.

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, 1995.

No context found.

H. Buxton and S. Gong. Advanced visual surveillance using bayesian networks. In Proc. International Conference on Computer Vision, 1995.

No context found.

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, pages 111--123, Cambridge, Massachusetts, June 1995.

No context found.

H. Buxton and S. Gong. Advanced Visual Surveillance using Bayesian Networks. In International Conference on Computer Vision, pages 111--123, Cambridge, Massachusetts, June 1995.

No context found.

Hilary Buxton and Shaogang Gong. Advanced visual surveillance using Bayesian networks. In Proceedings, IEEE Workshop on Context-Based Vision, 1995.

No context found.

H. Buxton and S. Gong, `Advanced Visual Surveillance using Bayesian Networks', in Proc. Workshop on Context--Based Vision (In Conjunction with Fifth International Conference on Computer Vision (ICCV95)), 19 June 1995.

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