UKF-assisted optimization for single-pixel imaging and target tracking

This study proposes a novel optimization framework integrating Unscented Kalman Filter (UKF) prediction with an Augmented Lagrange Method (ALM) to address the critical challenges of computational efficiency and nonlinear motion target tracking handling in single-pixel imaging (SPI). The algorithm exploits inter-frame correlation by using UKF-generated prior templates to initialize the ALM sparse reconstruction process, significantly accelerating convergence. Validation on 128×128 simulated UAV image sequences demonstrate 5× faster reconstruction (2.1ms per frame), 2.11 dB PSNR improvement than two-dimensional compressed sensing algorithm (V2DALM) under 25% sampling ratio. UKF tracking achieves high positional accuracy (RMSE