Fast Recursive Computation of Composite Correlation Filters

Abstract

Abstract—Algorithms for tracking multiple objects using correlation filters require calculation of a composite filter based on synthetic discriminant functions in order to increase the robustness to changes in posture, partial overlapping of objects by other objects, scaling, rotation, nonuniform illumination, and complex background. The calculation algorithm has large computational complexity. In this paper, we propose recursive calculation of a composite filter using algorithms for rapid inversion of matrices to accelerate synthesis of the filter. Computer simulation results on calculation of a composite correlation filter are presented and discussed in terms of the computation accuracy and the rate of calculation.