Sensor Placement Optimization Method for People Tracking

In this paper, we deal with a sensor placement optimization problem for multi-point pedestrian flow monitoring systems, and provide an efficient algorithm. Our goal is to build a realistic and accurate model of the monitoring systems' pedestrian flow estimation performance in terms of their estimation errors. Also our model can represent sensor placement considering the number, types, locations and capabilities of sensors in urban scenarios. To this goal, we formulate the sub-problem of estimating accuracy achieved by a given monitoring system under a given placement of sensors. Using a solver for this sub-problem as a sub-module, we also design an algorithm to determine the optimal sensor placement. This algorithm employs a simulated annealing (SA) based approach and iteratively improve solutions to converge to near optimal solutions. Through performance evaluation using our HumanS simulator [1], which simulates human detection sensors, pedestrian behavior and floor structures altogether, we have verified that a derived sensor placement by our proposed method could detect pedestrian flows with high accuracy for an underground city and its estimation error was about 1 percent.