A Sequential Bayesian Algorithm for Surveillance With Nonoverlapping Cameras

Abstract

Visual surveillance in wide areas (e.g. airports) relies on sparsely distributed cameras, that is, cameras that observe nonoverlapping scenes. In this setup, multiobject tracking requires reidentification of an object when it leaves one field of view, and later appears at some other. Although similar association problems are common for multiobject tracking scenarios, in the distributed case one has to cope with asynchronous observations and cannot assume smooth motion of the objects. In this paper, we propose a method for human indoor tracking. The method is based on a Dynamic Bayes Network (DBN) as a probabilistic model for the observations. The edges of the network define the correspondences between observations of the same object. Accordingly, we derive an approximate EM-like method for selecting the most likely structure of DBN and learning model parameters. The presented algorithm is tested on a collection of real-world observations gathered by a system of cameras in an office building.

A Review of Indoor Localization Technologies: towards Navigational Assistance for Topographical Disorientation

“…The articles selected for this section can be distinguished based on the sensor required to estimate the location of the tracked object. Some articles proposed the use of cameras to locate a mobile subject or device via image processing [58,5,94,24,21,93,97,89,39,70,31,48,10]. In contrast, some other articles present localization solutions based on non-image processing devices [15,56,55,14,33,90].…”

“…Instead, most of the articles presented a solution in which the tracking process was limited to the field of view of a single camera. Therefore, only articles [58,97,89,48,10] were considered as an indoor tracking solution based on our inclusion and exclusion criteria. In [58], the authors proposed a system that detected elements of the scene which were not part of the static environment.…”

“…The authors reported correct user recall values of 87.21% with an availability of 73.55%. A similar approach was presented in [97]. The authors used colour-based features assuming vertical differentiation of colour regions of a human figure while standing up.…”

Distributed EM Learning for Appearance Based Multi-Camera Tracking

“…The probabilistic model could be enhanced with a transition model, using discrete features, like camera index and wall clock time. We have planned to incorporate such a transition model into the algorithm along the ideas presented in [3,14].…”

Object Modeling for Multicamera Correspondence Using Fuzzy Region Color Adjacency Graphs

“…Some multicamera person tracking methods partition body to predefine number of segments and track color information of each independently. Reference [13] claims that color information of body, pelvis, and feet are the most stable information even in distributed cameras. In [14] authors cluster a person according to its color and motion features using watershed and K-means algorithm to recognize gestures and track each.…”