Surveillance cameras are an essential component of the overall crime prevention strategy and Law Enforcement agencies now heavily use imagery collected by surveillance cameras in solving crimes. Cameras installed on roadways are used to collect traffic data with the aim to better manage traffic, relieve congestion, respond to accidents, etc. Surveillance cameras are also used in high-stake public places, such as airport, train stations, metro stations and bus terminals. In recent years, cameras have also been used to record and prevent incidents of police brutality. In short, as a society, we are increasingly reliant on cameras.

In this paper, we consider the multi-camera tracking and the camera active control (pan and tilt). Auction mechanism from economics is developed to choose the best available camera. By modeling the camera bids with prior knowledge of the camera homographies, the system can "think" ahead to perform necessary panning or tilting operations. The uncertainties of homographies are considered inherently in the metrics used for computing camera bids. Further, to have a better tracking result, we use multiple trackers simultaneously. The trackers are rectified periodically based on the previous auction results. The proposed approach is evaluated in a realworld camera network.

Sensor networks have been a very active area of research in recent years. However, most of the sensors used in the development of these networks have been local and nonimaging sensors. The emerging development of video sensor networks poses its own set of unique challenges, including high-bandwidth and low latency requirements for real-time processing and control. This paper presents a systematic approach by detailing the design, implementation, and evaluation of a large-scale wireless camera network, suitable for a variety of practical real-time applications. We take into consideration issues related to hardware, software, control, architecture, network connectivity, performance evaluation, and data-processing strategies for the network. We also perform multiobjective optimization on settings such as video resolution and compression quality to provide insight into the performance trade-offs when configuring such a network and present lessons learned in the building and daily usage of the network.

In this paper, we apply the auction-based theories in economics to camera networks. We develop a set of auction protocols to do camera active control (pan/tilt/zoom) intelligently. Unlike the economic auction, the bid price in our case is formulated to have a vector representation, such that when a camera is available to follow multiple objects, we consider the "willingness" of this camera to track a particular object. Most of the computation is decentralized by computing the bid price locally while the final decision is made by a virtual auctioneer based on all the available bids, which is analogous to a real auction in economics. Thus, we can take the advantages of distributed/centralized computation and avoid their pitfalls. The experimental results show that the proposed approach is effective and efficient for dynamically active control based on user defined performance metrics.

Camera assignment and hand-off are some of the key image processing problems in a video network. In this paper, we propose a new approach for camera assignment and handoff in a video network. The camera assignment problem is modeled as a weakly acyclic game which allows the design of utility functions based on different user-supplied criteria. A theoretical and experimental comparison of the proposed approach with the two recently proposed approaches based on potential game theory and constraint satisfaction problem is provided. This comparison shows that the proposed approach is theoretically more general and computationally more efficient than the other approaches.

Sensor networks have been a very active area of research in recent years. However, most of the sensors used in the development of these networks have been local and nonimaging sensors such as acoustics, seismic, vibration, temperature, humidity, etc. The development of emerging video sensor networks poses its own set of unique challenges, including high bandwidth and low latency requirements for real-time processing and control. This paper presents a systematic approach for the design, implementation, and evaluation of a large-scale, softwarereconfigurable, wireless camera network, suitable for a variety of practical real-time applications. We take into consideration issues related to the hardware, software, control, architecture, network connectivity, performance evaluation, and data processing strategies for the network. We perform multi-objective optimization on settings such as video resolution and compression quality to provide insight into the performance trade-offs when configuring such a network.

Video networks are becoming increasingly important for solving many real-world problems. Multiple video sensors, usually cameras, require collaboration when performing various tasks. One of the most basic tasks is the tracking of objects, which requires mechanisms to select a camera for a certain object and hand-off this object from one camera to another so as to accomplish seamless tracking. In this paper, we provide a comprehensive comparison of current and emerging camera selection and hand-off techniques. We consider geometry, statistics, and game theory-based approaches and provide both theoretical end experimental comparison using centralized and distributed computational models. We provide simulation and experimental results using real data for various scenarios of a large number of cameras and objects for in-depth understanding of strengths and weaknesses of these techniques.

Unlike existing methods that used the human actions or trajectories to analyze the human activity in overlapping field-of-views, this paper proposes the appearance and travel time-based human activity classification in the camera network of non-overlapping field-of-views. The mixture of Gaussian-based appearance similarity model incorporates the appearance variance between different cameras to address changes in varying lighting conditions. To address the problem of limited labeled training data, we propose the use of semi-supervised Expectation-Maximization algorithm for activity classification. The human activities observed in a simulated camera network with nine cameras and twentyfive nodes are classified into one normal and three anomalous classes. A similar camera network is built and tested in real-life experiments, in which the proposed approach achieves satisfactory performance.