Research On Data Association And State Tracking In Multi-Target And Multi-Sensor System

With the development of technology, multi-target tracking system begins to play an important role in modern monitor system. And in order to achieve more accurate tracking of target state, it is also necessary to improving accurateness of measurement. For the tracking system, it is also essential to promote correct rate of data association, which is an important part and has a huge influence on state estimation and filtering. Further more, it is indispensable to analysis multi-sensor system, which could take the advantage of data redundancy and complementarity, and improve the precision of state estimation. At the same time, it could not be neglected that natures of sensors in tracking system might be quite difference, and with more noise, thus it may be quite difficult to processing and making use of those mearsuremnt effectively.After analyzing the advantage and disadvantage of multi-sensor and multi-target system, the procedure of tracking targets can be divided into several parts, including multi-target data association, estimation and updating of state, and multi-sensor track fusion. Each of those parts is indispensable to improving the tracking quality, and the article can be descrbed as follows:1) As most important part of multi-target tracking, data association has attracted lots of attentions, and several methods have been designed to solve it, such as the Nearest Neighbor and it modified. But those methods do not consider the influence of sensor bias, which would destroy the performance of tracking system. In order to handle this problem, a novel modified version of Nearest Neighbor, which consider the partial position between multiple targets, is presented. And experiments show that the modified version could achieve a better performance.2) Following data association, filtering and updating of target state begins to play an important role in tracking. After analyzing of Kalman Filtering and Extended Kalman Filtering, it is clear that those methods are not suitable for non-Gaussian system. In order to tracking in non-Gaussian system, particle filter is introduced to process the non-Gaussian noise, and the experiment shows that it could achieve an improvement of tracking. But those methods could be destroyed in the presence of sensor bias. In order to processing this problem, the sensor bias would be estimated using association strategy. Then, the updating of target takes into account pseudo measurement, which obtained through eliminating bias from measurement.3) With development of sensor technology, multi-sensor tracking system begins to be used widely. Compared with single sensor system, multi-sensor system could obtain more information about state. In order to make use of those measurements, the distributed processing architecture was used widely in tracking system. The article presents some track fusion methods for multiple sensors under distributed architecture, and finds that with weighted fusion method of track, the system can achieve target tracking with less error.