【作者】 康健；

【导师】 谢红；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2013， 博士

【摘要】 多传感器信息融合技术是国家重点科研项目,近年来,世界各国都投入了大量的人力、物力来对多源信息融合技术进行理论和应用方面的研究,目前该新技术主要用于军事领域,民用前景也十分广泛,可见该技术的重要性。本文主要围绕多传感器信息融合技术中的一些关键技术展开研究,论文的主要研究内容包括解决数据预处理技术中的野值剔除、数据关联、数据决策以及多传感器信息融合的实际应用。首先,数据预处理技术是提高融合系统精确度的前提,由于噪声等因素的干扰,导致传感器接收到的数据精确度不高,甚至会出现偏差严重的数据。针对这一问题,提出了基于新息变化的野值检测方法,该方法考虑新息的变化情况来对野值进行检测,利用卡尔曼滤波获得的新息情况实时的对量测是否为野值进行判断,并通过加权函数计算量测的权重用来对野值点进行数据补偿来解决野值问题,以此提高数据预处理部分数据的精度。通过仿真证明了算法的有效性。其次,对于数据关联算法的研究部分,针对在高杂波密度环境下的单目标跟踪算法精度不高的现象,提出了基于证据理论的概率数据关联算法,该方法充分利用传感器的量测信息和通过概率关联算法获得的状态估计信息,并通过改进的证据理论合成算法对信息进行融合,提高了目标的跟踪精度。对于多杂波环境下多目标的跟踪问题,在单目标跟踪算法的基础上进行了扩展研究,提出了基于证据理论的联合概率数据关联算法,该方法有效解决了多杂波、多目标情况下,经典数据关联算法目标跟踪精度过差的问题。此外,在提高多目标跟踪精度的基础上,为了减少计算量,提高目标跟踪的实时性,提出了基于最大模糊熵的数据关联改进算法,该方法利用最大模糊熵来对跟踪门内的量测进行重新分配,解决了随着目标数目的增多,可行性矩阵成几何倍数增长的缺陷,减小了计算量。并通过仿真实验验证了算法的性能。再次,对信息融合技术中数据决策部分的相关算法受限于先验知识以及不能够有效处理不确定性信息的问题,提出了针对冲突的改进DS(Dempster Shafer)证据理论算法。算法通过分析证据的一致性和确定焦元的重要性两方面入手,解决了DS证据理论存在的一票否决现象和证据冲突过大的问题,并降低了判决结果的不确定性。对于需要考虑多传感器置信度的决策问题,提出了基于传感器信任度的DS证据理论改进算法,利用灰关联获得传感器的置信度,并结合传感器获得的焦元信息和传感器置信度综合对目标进行判决,理论分析和实验仿真均表明算法具有良好的判决效果。最后,在多传感器信息融合的应用问题中,针对于同类传感器的信息融合,提出了改进的多传感器卡尔曼滤波的融合方法,利用提出的DS证据理论在权值分配上的改进方法对传感器接收的量测信息进行融合处理来得到更加准确的融合信息。并对雷达和红外的异类传感器的信息融合系统进行了仿真,仿真结果表明融合后获得了更高的精度。更多还原

【Abstract】 Multi-sensor information fusion technology is the important research project of nation, in recent years, many countries around the world have invested lots of manpower and material in order to researching the multi-source information fusion theory and its application. At present, not only the new technology is mainly used in military field. but also its civilian prospects are very widespread. Thus, we can know the importance of this technology. This article mainly researches some key parts of the multi-sensor information fusion technology, the main work of the paper include outliers elimination, data association, data decision in the data preprocessing technology and the applications of multi-sensor information fusion technology.Firstly, the data preprocessing technology is the premise of improving the fusion system’s accuracy. Due to some factors such as noise interference, the data received by the sensor does not have high accuracy, even some data has serious deviation.Concerning this issue, the paper proposes a kind of outliers detection method based on the changes of innovations. The method detects outliers according to the changes of innovations and utilizes Kalman filter to obtain innovations in order to judge timely whether the measurement is outlier. Meanwhile, the paper solves the problem of outliers through compensating the data points for outliers and the basis of compensation is calculating measurement weight through the weighing function. As a result, the accuracy of the data preprocessing part gets improved. The simulation results show this algorithm is effective.Secondly, This paper focuses on the data association technology. Aiming at the problem of single target tracking owns low accuracy in clutter, the paper proposes the probabilistic data association algorithm based on evidence theory. The algorithm utilizes measurements of sensor and state estimates calculated by probabilistic data association algorithm, then fuses the information with improved evidence theory synthesis algorithm. As a result, the target tracking accuracy gets improved. For multi-target tracking problem in clutter,the paper maks further research on the basic of single target tracking, proposes the joint probabilistic data association algorithm based on evidence theory. The classical data association algorithm for target tracking has poor accuracy at the situation of multi-target and clutter environment. The problem gets solved by the algorithm of this paper.Besides, on the basic of improving the multi-target tracking accuracy, in order to reducing the amount of calculation and improving the real-time of target tracking, this paper proposes the improved data association algorithm based on the maximum fuzzy entropy.The algorithm utilizes the maximum fuzzy entropy to renewedly distribute measurements that are in the tracking gate. Feasibility matrix grows by geometric multiples as the number of goals increases. The proposed algorithm can solve the above problem.At the same time, the algorithm reduces the amount of calculation.The simulation results show the superiority of the proposed algorithm.Thirdly, in data fusion technology, the related algorithms about data decision are limited by priori knowledge and the problem of lacking the ability of dealing with uncertainty information.In this paper, according to the conflict, an improved algorithm based on DS evidence theory is proposed. By analyzing the consistency of the evidence and the importance of determining focus information, the paper solves the existing problem about one-veto veto and the overlarge evidence conflict.It also reduces the uncertainty of judgment result.For the decision problems which need considering the multi-sensor confidence, the paper proposes the improved DS evidence theory algorithm based on confidence of sensors.The method gets the confidence of sensors with grey correlation, then judges the target with the focuss information of sensors and confidence of sensors.Theoretical analysis and experimental simulation indicates that algorithm has a good judgment effect.Finally, in the applications of multi-sensor information fusion, for the information fusion of similar sensors, the paper proposes an improved multi-sensor Kalman filter fusion algorithm. The improved algorithm is based on DS evidence theory in the distribution of weights. Through the algorithm, we can get more accurate fusion information according to processing the information received by sensors. And the paper makes simulations about heterogeneous sensor information fusion systems of radar and infrared, the simulation results show that the method gets higher precision after fusion.更多还原