Adaptive recursive optimized extrinsic self-calibration in distributed visual sensor networks

The extrinsic calibration of cameras in distributed visual sensor networks (VSNs) is indispensable for many applications such as intelligent surveillance, environmental and traffic monitoring. This paper proposes an adaptive recursive optimization framework for self-calibrating the extrinsic parameters of the nodes in distributed VSNs. First, the proposed approach utilizes the observations of a moving pedestrian to eliminate the precise auxiliary patterns. Second, a novel distributed algorithm is presented to reduce the amount of the data transmission and improve the scalability. In this algorithm, each visual sensor node calibrates itself into the local World Coordinate System (lWCS) independently by estimating the feet-head homology. And all lWCSs are registered into the World Coordinate System through an automatic planar trajectory alignment algorithm. Finally, aimed at improving the calibration accuracy and avoiding the local minimization, an adaptive chaos particle swarm optimization method is proposed to minimize the re-projection and trajectory matching error recursively. The performance of experimental results shows the proposed method could achieve comparable calibration accuracy to the centralized method with significantly less data transmission, and gain better robustness than the Levenberg-Marquardt optimized method.