Distributed camera control via moving horizon Bayesian optimization

Abstract

Active distributed optimal control of PTZ (pan, tilt, zoom) camera networks can yield accurate target tracking with high resolution imagery at opportunistic time instants, even when the number of targets exceeds the number of cameras. However, one step ahead methods have limited ability to simultaneously tradeoff competing objectives, potentially sacrificing smoothness of the PTZ trajectory. In addition to mechanical wear, this causes video feed motion blur, making it unsuitable for analysis. To address this challenge, we use a distributed optimization algorithm over a moving horizon with target tracking and PTZ smoothness constraints. The planning horizon enables the immediate actions of the camera control module to consider future effects. The solution approach is designed using a Bayesian formulation within a game-theoretic framework. The Bayesian formulation enables automatic trading-off of objective maximization versus the risk of losing track of any target, while the game-theoretic design allows the global problem to be decoupled into local problems at each camera. The feasible PTZ parameter set is defined by constraints on target tracking performance and PTZ smoothness. Cameras alter their own PTZ sequences by using information received from neighboring cameras, and broadcast the updated sequences to their neighbors. This article presents the theoretical solution along with simulation results.