| Title |
Distributed Particle Filter based Vehicle Tracking System with α-stable Parameter |
| Authors |
박재성(Jaesung Park) ; 윤창용(Changyong Yoon) |
| DOI |
https://doi.org/10.5370/KIEE.2020.69.1.190 |
| Keywords |
Distributed Particle Filtering; Object Tracking; -stable Distribution; Gaussian Mixture Model |
| Abstract |
This paper proposes a system with multiple sensors to track vehicles in real time while driving. Environmental information on real-time roads has the characteristics of nonlinear, non-gaussian distributions, and the performance of tracking systems tends to decrease if sudden changes occur and outliers are included. To overcome these shortcomings from vehicle tracking systems, we proposes algorithms that maintain stable and robust characteristics in real-time road environments using distributed particle filters and symmetrical-stability parameters. The distributed particle filter adopts a Gaussian mixed model exchanging parameters between adjacent sensor nodes to approximate the posterior distribution of the weighted particles. The average consensus filter is used to enable each sensor node to interact with the neighboring node. Also, outliers such as impulse, the state estimation method applied to the particle filter-based tracking system is likely to result in degraded performance. To resolve these problems, this paper proposes the method to use a probability density function approximated by the particles generated, as using -stability distribution value appropriate to the circumstances. The experimental results show that the proposed method has better performance than other traditional particle methods, even when multiple sensors are used to detect multiple objects in nonlinear environment where rapid changes occur. |