Self-calibration and Cooperative State ...

In this paper we present an algorithm, which enables distributed

visual sensor networks to autonomously calibrate the network and

dynamically build clusters to achieve cooperative object tracking

based on state estimation. A main focus is thereby on resource-

awareness and -efficiency, since we aim for low-power embedded

smart camera networks. We do not require any human interven-

tion or a-priori information about the network topology to achieve

calibration and tracking. Camera nodes first estimate relative posi-

tions and orientations and then use the common coordinate system

to enable cooperative state estimation. For that purpose, cameras

dynamically build clusters depending on their available resources.

New nodes joining the network are discovered and failing nodes do

not prevent others from their tasks. Compared to other methods,

our approach is not only able to handle sensor measurement er-

rors but also faulty camera positions gathered during the network

calibration process.