Cognitive visual tracking and camera control

Abstract

Cognitive visual tracking is the process of observing and understanding the behaviour of a moving person. This paper presents an efficient solution to extract, in real-time, high-level information from an observed scene, and generate the most appropriate commands for a set of pan-tilt-zoom (PTZ) cameras in a surveillance scenario. Such a high-level feedback control loop, which is the main novelty of our work, will serve to reduce uncertainties in the observed scene and to maximize the amount of information extracted from it. It is implemented with a distributed camera system using SQL tables as virtual communication channels, and Situation Graph Trees for knowledge representation, inference and high-level camera control. A set of experiments in a surveillance scenario show the effectiveness of our approach and its potential for real applications of cognitive vision.

Multi-Agent Autonomous Surveillance: A Framework Based on Stochastic Reachability and Hierarchical Task Allocation

“…One is associated with solving the reachability DP defined in Eq. (7) and the other with selecting the optimal allocation of objectives between cameras by solving the combinatorial optimization Problem (9). In general, both problems are very difficult and the complexity of obtaining the optimal solution scales exponentially in the number of cameras and targets, respectively.…”

“…The choice of the best allocation (see Problem (9)) results in the comparison of at most m 2 =4 ð Þþm ½ d possible solutions which is O m 2d À Á . Using the proposed task allocation mechanism we replace the calculation of the 4d-dimensional DPs defined for each surveillance objective to a collection of four-dimensional ones, followed by a search over Oðm 2d Þ possible task allocations.…”

“…Most approaches address the surveillance problem using a combination of static and PTZ cameras where feedback from static cameras is used to position the moving ones (see, for example, Refs. [5][6][7][8][9][10]). To reduce the complexity on large camera networks, several of these works focus on distributed implementations and in some cases relax the requirement of static cameras (see, for example, [10][11][12]).…”

“…Our experimental setup is similar to Refs. [5][6][7][8][9] since we are using a global eye camera, but our framework can be used with any type of global position sensor. We dedicate a complete section to elaborate on the implementation details of the proposed approach and carry out a series of experiments to analyze the overall performance.…”

Feature-based automatic configuration of semi-stationary multi-camera components

“…tracking of persons with different cameras and getting detailed view of the person for identification where it is possible as shown in Ref. 4. A rough overview of an entire scene with a wide-angle camera is not enough: On the one hand a situation recognition system needs enough visual input to detect fine grained situations, and on the other hand when recognizing the occurrence of a car or the presence of a person a close-up view of that interesting agent is required to further refine the information or to identify this potentially interesting agent.…”