Target Fusion Tracking Technology And Performance Prediction

In this paper, the target tracking algorithms and their performance evaluation techniques are studied under non-linear measurement and high-density clutter circumstance.Firstly, the paper introduces in brief the research subjects, reviews the predecessors' achievement, summaries the current situation and the existing problems in target tracking field. Afterwards, the main subjects and research background of this essay are expatiated.Aimed at target tracking under non-linear measurement, this paper extends two-dimension CMKF algorithms to three-dimension, uses canonical transform to obtain decoupled CMKF algorithm which makes the theoretical analysis for the algorithm easily. In order to improve the tracking precision at the beginning, an approach for estimate initial value of CMKF and decoupled CMKF algorithm are proposed. In the meantime, the canonical transform technique is generalized and adopted to solve decoupling problem of maneuver target "current" statistic model, and the transformation matrix is derived for the 2D radar measurement. Lastly, the concept of "Transient Steady State " (TSS) and its realized approach is put forward, and TSS filter's error variance is obtained, which provides theoretical reference for the analysis of the tracking performance under the non-linear measurements.Given the condition of the fusion tracking under multiple sensors non-linear measurements, this paper applies the conclusion of the analysis of single sensor decoupled CMKF, TSS filter and TSS variance to the fusion tracking system. It also studies their application in basic measurement fusion and track fusion algorithm, and covariance recursive formula of track fusion is deduced. Which offers a new way for steady-state performance evaluation of multi-radar fusion tracking.Under the background of target tracking in high-density clutter, the calculation formula of the PDA algorithm's error variance is analyzed and revised firstly, and the modified PDA (MPDA) is derived, which not only improves the tracking performance, but also makes the error variance and the real error of the algorithm match well. Then, two methods are adopted to evaluate and to predict its performance. One is the steady-state performance evaluation based on the Riccati equation. The same conclusion as derived from the original PDA under approximate condition is concluded this way without any approximations. The second method is the instant-state performance prediction based on the HYCA method. This method not only gives the off-line recursive error variance relation, but also gets a series of performance measurement such as track life.On the ground of MPDA algorithm and the conclusion of its steady-state performanceevaluation, the choice of the detection threshold becomes a matter of optimization under the condition of properly chosen detection model and hypothesis. And the analytic expression of auto-adjusted detection threshold can be deduced via approximate fitting attenuation factor derived from the conclusion of performance estimation. It presents a novel approach for the optimization of detection-tracking system.