A Review of Multi-source Trajectory Data Association for Marine Targets
LU Qiang;WU Lin;CHEN Zhao;WANG Qi;XU Yongjun;KAN Rongcai;University of Chinese Academy of Sciences;Special Technology Research Center, Institute of Computing Technology,Chinese Academy of Sciences;Guangdong Key Laboratory of Big Data Analysis and Processing;Troops 92896;
With the globalization of the Belt and Road national strategy, the volume of shipping trade is increasing rapidly. As a result, the problem of the safety of maritime navigation and monitoring has become increasingly prominent. The real-time monitoring of large-scale ships, based on the spatio-temporal data, through target tracking and information fusion is an effective method, but it also faces great challenges. Data association,as the basis and a key step of target tracking and information fusion, has important application value in military and civil fields. This paper summarizes the problems related to data association. Firstly, the data sources for trajectories of the marine targets were introduced and compared, showing its necessity and feasibility. Then two kinds of problems in data association, i.e., measurement-to-track association(MTTA) and track-to-track association(TTTA), were described. Based on the data association methods in MTTA, we abstracted a data association model consisting of state estimation and association judgment, and described the Kalman filter used generally in state estimation. After that, the basic principles and improvements of nearest neighbor(NN),probabilistic data association(PDA), joint probabilistic data association(JPDA) and multiple hypothesis tracking(MHT)were introduced. NN implements the data association using the distance between the measured and predicted values. PDA, considering only a single target, calculates the association probability of each measurement in the circumstance with presence of clutter and target missing, and associates the measurement with the maximum association probability to the target. JPDA as the extension of PDA, suitable for multiple targets,calculates the joint association probability of measurements and targets by joining all targets, and selects the association event corresponding to the maximum joint association probability as the association result. MHT is a multi-scan multi-hypothesis method and has the characteristics of track creation, maintenance, deletion and false alarm.It achieves the optimum in theory by maintaining multiple possible hypotheses generated by each association cycle. The key to the MHT is how to control the scale of the hypotheses by effective pruning in order to improve the efficiency of time and space of the algorithm. With regard to TTTA, two kinds of methods, based on statistics and fuzzy mathematics, were introduced respectively. The statistics methods consist of NN/K-NN/MK-NN, double threshold track correlation, sequential track correlation, etc. The key of fuzzy methods is the construction of fuzzy factor set and membership function. We also introduced the evaluation methods for data association. Finally, the problems in the existing methods, e.g., the application scenarios, and further researches were explained.
【Fund】： 中国科学院重点部署项目(ZDRW-ZS-2016-6-3);; 广东省大数据分析与处理重点实验室开放基金项目(201804)~~
【CateGory Index】： U675.79
【CateGory Index】： U675.79