Coopetitive visual surveillance using model predictive control

Abstract

Active cooperative sensing with multiple sensors is being actively researched in visual surveillance. However, active cooperative sensing often suffers from the delay in information exchange among the sensors and also from sensor reaction delays. This is because simplistic control strategies like Proportional Integral Differential (PID), that do not employ the look-ahead strategy, often fail to counterbalance these delays at real time. Hence, there is a need for more sophisticated interaction and control mechanisms that can overcome the delay problems. In this paper, we propose a coopetitive framework using Model Predictive Control (MPC) which allows the sensors to not only 'compete' as well as 'cooperate' with each other to perform the designated task in the best possible manner but also to dynamically swap their roles and sub-goals rather than just the parameters. MPC is used as a feedback control mechanism to allow sensors to react not only based on past observations but also on possible future events. We demonstrate the utility of our framework in a dual camera surveillance setup with the goal of capturing the high resolution images of intruders in the surveyed rectangular area e.g. an ATM lobby or a museum. The results are promising and clearly establish the efficacy of coopetition as an effective form of interaction between sensors and MPC as a superior feedback mechanism than the PID.