Research and Application of Moving Target Tracking Based on Multi-innovation Kalman Filter Algorithm
Moving objects tracking is an active problem in computer vision and has a wide variety of applications. The Kalman Filter algorithm has been commonly used for estimation and prediction of the target position in target tracking domain, of which the algorit
- PDF / 667,956 Bytes
- 16 Pages / 439.37 x 666.14 pts Page_size
- 15 Downloads / 232 Views
Abstract. Moving objects tracking is an active problem in computer vision and has a wide variety of applications. The Kalman Filter algorithm has been commonly used for estimation and prediction of the target position in target tracking domain, of which the algorithm is adaptive to linear system, but the error of Kalman Filter will become large or even diverging when the target status changes suddenly. In this paper, multi-innovation theory is applied to target tracking, and the Multi-innovation Kalman Filter is proposed. Multi-innovation Kalman Filter has better precision and stability, because Multi-innovation Kalman Filter takes not only the moving targets’ current state of motion into consideration, but also the time before. In addition, the authors theoretically analyzed the convergence of improved Multi-innovation Kalman Filter algorithm. Finally, simulation results show that the improved algorithm Multi- innovation Kalman Filter is superior to the traditional Kalman Filter. Keywords: Target tracking · Kalman filter · Multi-innovation · Multi-innovation Kalman Filter · Simulation analyses
1
Introduction
With the development of computer technology, computer capacity has been greatly improved, and the target tracking has become one of the hottest topics at home and abroad [1]. It has been importantly and widely application in the fields of civilian, military, transportation and others [2, 3]. But how to accomplish correct and fast target tracking and how to reach good real-time performance and robustness are the key problems to be solved. KF algorithm [4, 5] is widely used for target tracking, because KF can achieve the optimal estimation and better results of target tracking if the system equation, system noise and observation noise are all known [6, 7]. However, generally, the motion state of the observed object keeps a time-varying motion, instead of a uniform linear one. In this case, traditional KF predictive algorithm loses its superiority in target tracking, the tracking precision declines and the rate of convergence slows down. It may even result in losing track of the object. © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 90–105, 2015. DOI: 10.1007/978-3-662-48570-5_10
Research and Application of Moving Target Tracking
91
Considering the above problems of traditional KF, we generally use the adaptive filtering technique to solve them [8-12]. The authors of [8,9] come up with an improved adaptive KF algorithm, which introduces the forgetting factors based on fading memory index weighting, and gets the best forgetting factor by the method of forecasting residual error. At the same time, it also takes measures to ensure the semipositive definiteness and positive definiteness of the noise estimation variance matrix and the measurement noise estimation variance matrix, thereby avoids the filter divergence. In [10], the method of KF target tracking, based on genetic algorithm, is proposed. The KF prediction is used to determine the candidate regions of t
Data Loading...
