Nonlinear Pedestrian Tracking by Unscented Particle Filter for Autonomous Vehicle Under Existence of Non-Gaussian Distributed Uncertainty

This paper proposes an Unscented Particle Filter(UPF) based pedestrian tracking technique in process and measurement systems with non-Gaussian uncertainty distribution. Accurate recognition and state estimation of pedestrian objects are required for successful Autonomous Driving (AD). Thus, we performed modeling to estimate the nonlinear relative movement of a pedestrian and designed UPF. In order to confirm the performance and effectiveness of UPF applied in this study, multiple simulation experiments were conducted under various uncertainty scenario combinations. In particular, the performance of nonlinear state estimation was compared under the non-Gaussian distribution characteristic conditions that both the object tracking model and recognition by extroverted sensors for AD can have. As a result, the proposed UPF-based tracking method for all results has the slightest error compared to other baseline nonlinear filter methods.