Multi-target Tracking With State-dependent Clutter

In a multi-target tracking problem, clutter is often modeled as a Poisson point process. Therefore, it can be described by the intensity information. Clutter intensity contains the average number of clutters per frame and the probability density of the clutter spatial distribution. Normally, the probability density distribution is known as a priori, and has an assumption that the clutter uniformly distributes within the observation area. But in the reality, the clutter intensity can be changed with the difference in observations, electromagnetic interferences or other factors. For most sensors, the observed distribution of clutters is more likely in the vicinity of targets and the clutter intensity may be complex and unknown. If the actual clutter intensity is strikingly different from the assumption, the tracking performance of a conventional filter will be degraded. At same time, when the amount of clutter measurements increase, the time complexity of filtering algorithms will increased, and its real time performance must be affected.In view of the above problems, this thesis proposes the improved GM-PHD filtering and GM-CPHD filtering algorithm on the research progress of Gaussian mixture PHD filtering. The specific contents of this paper include:(1) GM-PHD filter with state-dependent clutter. In traditional filtering methods, clutters are often assumed to obey a uniform distribution in the entire monitoring area. However, for many sensors, clutters may concentrate in target-containing regions. Under this condition, the performance of the traditional multi-target tracking filter can be degraded. Aiming at solving this problem, this thesis proposes an improved algorithm based on Gaussian Mixture probability hypothesis density(GM-PHD) filter to deal with state-dependent clutters. First, the relationship between state and clutters is modeled by using the uniform distribution centered on the target state. Then, the clutter intensity is calculated according to the distribution of clutters in the whole monitoring area and is applied to updating the filter. The simulation results show that the improved filter can track targets’ trajectories more effectively in the environment of state-dependent clutters than the standard GM-PHD filter.(2) Fast GM-PHD filter for multi-target tracking. In the traditional GM-PHD filter, all measurements received at current time are used to update different types of targets. Much time is spent on updating targets because of using invalid measurements. A kind of fast multi-target tracking filter is proposed in this paper. Firstly, Gaussian components are divided into two parts. One part is birth targets and the other is survival targets. Then the residuals between survival targets and all measurements are calculated. Next, only the measurements which fall in the elliptical gate are used to update survival targets. Similarly, the residuals between birth targets and remaining measurements are calculated, and only those measurements which fall in the elliptical gate are used to update birth targets. In this way, we could minimize invalid measurements and reduce the computing complexity. The experimental results show that the new method not only reduces the time complexity greatly but also insures the accuracy of target tracking. Its performance is better than the traditional GM-PHD filter as a whole.(3) A GM-CPHD filter with state-dependent clutter. In practical application, for many sensors the observed clutter is easier to concentrate in the target area, namely the state-dependent clutter, which is different from the known clutter distribution in traditional filtering algorithm, so the accuracy of traditional multi-target tracking and real-time performance of algorithm will be greatly affected. To solve this problem, this paper proposes a kind of GM-CPHD filtering algorithm under state-dependent conditions. First, the relationship between clutters and state is modeled. Then, recalculated clutter intensity according to the distribution of clutters in the whole surveillance area, which applied to update process; at the same time, in order to reduce the time complexity of the proposed GM-CPHD filter, an adaptive gating strategy is adopted to make a pretreatment for measurement, then the measurements which fall into the threshold will be used in update step. The simulation results show that in the environment of state-dependent clutters, the proposed algorithm has better filtering accuracy and lower time complexity than traditional algorithm, and meet the requirement of real-time very well.