A new filtering strategy for target tracking application using the second form of Smooth Variable Structure Filter

Abstract

Target tracking process consists of tracking and measuring a quantity over time, which is a difficult estimation problem. The latter gives erroneous measurements that are affected by the unknown number of targets to be tracked, as well as, measurements that are corrupted by noise. Thus, overcoming the target tracking problem includes basically solving the filtering problem which leads to a data association problem. Tracking problems that force lightly nonlinear models are solved appropriately with the Extended Kalman Filter. Although, problems that use nonlinear models, applying analytical expressions of the Jacobian of dynamic systems, or measurement vector functions are potentially solved by the Unscented Kalman Filter. The first technique is well built, but not precise, while the second technique yields very accurate results, but in terms of complexity, it is not suitable for real-time applications. Another filter called Particle Filter used a Monte Carlo method based on probability distribution over the state is proposed. Subsequently, we used a Gaussian Mixture Model which is based on the Expectation Maximization algorithm for fitting mixture-of-Gaussian models is considered to guarantee the estimation of optimal parameters set for each target. In order to achieve better trade between optimality and robustness of the tracking effect, the Smooth Variable Structure Filter is proposed as an alternative. This latter filter does not make any assumptions about noise characteristics and provides accurate estimation results. Furthermore, in this article, we propose a new target tracking algorithm based on the second form of Smooth Variable Structure Filter, which is implemented in a simulation using different scenarios. Experimental scenarios, under realistic conditions, are presented and the obtained results confirm the effectiveness of the Smooth Variable Structure Filter approach compared to other filters.