A SIMULATED ANNEALING ALGORITHM FOR DETECTING MOVING TARGETS

Abstract

Target tracking and monitoring plays a crucial role in the intelligence collection domain. With the advancement of intelligence collection and data analysis methods, we can sometimes obtain a target’s initial and end locations of its desired trajectory, albeit with some uncertainty. Based on such intelligence information, the target's movement can be modeled as a stochastic process using a Brownian bridge, and the target’s geographical location probability distribution in time can be aggregated and mapped as a two-dimensional temporal heat map. Based on this model, we search for sensor deployment strategies that maximize the probability of target detection. This thesis adopts a random search method called simulated annealing and customizes it to the unique setting of target tracking to obtain a sensor configuration that approximately maximizes the target detection probability, accounting for uncertainty in intelligence information. To evaluate the performance of the proposed method, we perform an experimental design and compare the results from simulated annealing with a simple heuristic. Based on a drug trafficking scenario, we attempt to find the approximate best sensor configuration to maximize the probability the sensors successfully observing the target, given limited sensor coverage and uncertain intelligence.