Distributed data association in smart camera networks using belief propagation

Abstract

One of the fundamental requirements for visual surveillance with smart camera networks is the correct association of camera's observations with the tracks of objects under tracking. Most of the current systems work in a centralized manner in that the observations on all cameras need to be transmitted to a central server where some data association algorithm is running. Recently some works have been shown for distributed data association based solely on appearance observation. However, how to perform distributed association inference using both appearance and spatio-temporal information is still unclear. In this article, we present a novel method for estimating the posterior distribution of the label of each observation, indicating which of the objects it comes from, based on belief propagation between neighboring cameras. We develop distributed forward and backward inference algorithms for online and offline application, respectively, and further extend the algorithms to the case of unreliable detection. We also incorporate the proposed inference algorithms into distributed EM framework to simultaneously solve the problem of data association and appearance model learning in a completely distributed manner. The proposed method is verified on artificial data and on real world observations collected by a camera networks in an office building.

Distributed data association in smart camera network via dual decomposition

“…To overcome these difficulties, in [12] the authors proposed a spatio-temporal tree to model the dependence structure of the involved variables and use belief propagation algorithm for calculate the posterior distribution of each labeling variables in a distributed manner. The main limitation of [12] is that it requires the knowledge of the number of objects under tracking, which is usually unavailable in practice. In a recent work [32], the authors proposed a distributed Bayesian framework in which both the sampling space and the posterior distribution of labeling variables are inferred online, based on knowledge propagation between neighboring cameras.…”

“…One of the fundamental prerequisites for achieving these goals is the correct reconstruction of camera-to-camera trajectory of each object, or equivalently, grouping observations originated from the same object into a single track, which may be generated by different cameras at different time instants. This problem is often referred to as data association in camera networks [1,12], trajectory recovery [13], or camera-to-camera tracking [14][15][16][17][18]. In this paper, we assume that the detection and tracking problem within a single camera view has been solved, and we call the collection of quantities summarizing the features of tracked object as a virtual "observation", see Fig.2 as an example.…”

“…The problem of data association in camera networks can be treated in a Bayesian framework [1,12], in which each observation is assigned with a labeling variable, and the posterior distribution of each labeling variables are inferred based on all appearance and spatio-temporal evidence made in the whole networks. The resulting marginal distribution of labeling variable contains the complete knowledge about which object the observation came from and the trajectory of moving object can be recovered directly using these knowledge.…”

Fault-tolerant tracking of multiple targets in collaborative Camera Networks

“…1 for camera c 3 . We assume that camera c i can consistently identify each target within F i by employing appropriate distributed data association mechanisms; see [12] and references therein. Subsequently, it indicates the corresponding Z i,m where target t j lies.…”

Sensor management for enhanced catalogue maintenance of resident space objects

“…Autonomous control of surveillance sensors has recently been employed to improve videobased security systems [99][100][101]. This research describes a system that coordinates the steering and zooming of cameras to maximise coverage of a surveillance zone while minimising the FOV encompassing a target to improve feature detection.…”