Improving the Targets’ Trajectories Estimated by an Automotive RADAR Sensor Using Polynomial Fitting

A way to reduce the uncertainty at output of a Kalman filter embedded into tracker connected an automotive RADAR sensor consists adaptive selection parameters during tracking process. Different informed strategies for automatically tuning tracker’s and jointly learn state/output sequence using: expectation maximization; optimization approaches, including simplex algorithm; coordinate descent; genetic algorithms; nonlinear programming using finite differencing estimate gradient; Bayesian reinforcement learning; hyper-parameters in least squares, were already proposed. We develop here different semi-blind post-processing approach, which is faster more robust. Starting from conjecture that trajectory polynomial Cartesian coordinates, our method supposes fit data obtained polynomial. highlight, by simulations, improvement estimated trajectory’s accuracy fitting single multiple targets. propose new based on wavelets two steps: denoising part extraction, compares favorably with classical method. The effect proposed methods visible, targets’ trajectories estimations being hardly increased.