Sensor Scheduling for Target Tracking by Suboptimal Algorithms

We analyze the problem of tracking a single target, from which measurements are taken using noisy sensors. Each measurement is associated with the measurement error, usage cost, and physical and computational constraints. As it is acceptable for many real applications that target dynamics are modeled as a linear system that is impaired by white Gaussian noise. Moreover, it is assumed that sensor measurements are linearly distributed with white Gaussian noise. Optimal sensor scheduling is achieved by finding the sensor sequence that minimizes the total cost, which consists of the measurement error and sensor usage cost for the entire time horizon subject to specific system constraints. To handle this discrete optimization problem, we propose well-performing suboptimal methods for different energy constraints in sensor nodes. First, we propose a suboptimal method called the best step look-ahead technique, which performs very well when the energy constraints can safely be removed due to their negligible influence on the overall system. We also show that, under certain assumptions, the Viterbi algorithm can be applied as a suboptimal method to obtain attractive results. Second, the energy constraints are relaxed using Lagrangian multipliers to formulate the problem as a min-max optimization problem. We use particle swarm optimization to tune the Lagrangian multipliers and Viterbi algorithm to find the optimal sensor sequence. To illustrate the effectiveness of our algorithms in realistic settings, we study a numerical problem of single target tracking with several noisy sensors and convincingly show that the proposed methods perform better than existing methods.