移动目标活动规律挖掘方法研究与设计

发布时间：2018-03-24 16:38

本文选题：移动目标　切入点：时空同现模式　出处：《北方工业大学》2017年硕士论文

【摘要】：在现有目标跟踪技术的高速发展中,移动目标如行人、车辆和轮船等的轨迹时空数据可以被有效记录,这些时空数据中蕴含着大量潜在的有价值的模式及知识,这些潜在的模式在城市规划、国防军事、基于位置的服务等方面具有非常重要的研究价值。为发现移动目标的活动规律,本文采用同现模式分析方法进行共现分析,并在此基础上分析共现模式的部分周期性,本文的主要研究内容如下。针对如何从时空数据中发现有效模式的提取问题,本文提出了基于双层网络的时空同现模式挖掘算法。从行人、车辆和轮船移动目标的时空轨迹数据中发现时空同现模式需要计算时空兴趣度,为了简化时空兴趣度的计算方式,本文根据时空数据的特性提出双层时空网络建模方法,该时空网络有效保存了时空对象及时空元素之间的时空关系,在计算时空兴趣度时可以快速计算出时空频繁度。根据元素网络层可以减少大量冗余候选模式,从而减少计算量并降低内存空间的开销。本文在时空网络的基础上提出了基于时空网络的同现模式挖掘算法,该算法考虑了元素的有效周期,提出能表征模式的时空频繁度的权重特征值计量度,并采用模式链保存了全部时空同现模式集。实验表明,采用相同数据集并获取相同结果时,该算法相比Celik的算法及Wang等的算法运行效率较高。针对移动目标活动的部分周期性问题,本文提出移动目标部分周期性共现模式自适应挖掘算法。行人、车辆及轮船三类移动目标的共现活动通常具有部分周期性,本文将部分周期性模式分析应用在移动目标的共现规律研究中,提出部分周期性共现模式自适应挖掘算法。该算法依据元素有效周期加入了模式有效度,改进了共现频率的计算方式,算法的自适应体现在周期跨度及置信参数的确定中,依据时空数据的时间框架及初始化给出了周期跨度自适应确定方法,并依据最大周期跨度及保留全部部分周期性共现模式的原则,给出了置信参数自适应确定方法,然后根据周期先验性质,先判定较长模式的部分周期性再考虑子集模式的部分周期性,减少了共现分析中的共现频率计算,实验表明,本文所提的算法与Apriori-like算法及Naive算法相比,能自适应准确计算出周期跨度及置信参数,并保留了全部的部分周期性共现模式,提高了挖掘模式的运行效率。
[Abstract]:With the rapid development of the existing target tracking technology, the track time and space data of moving targets such as pedestrians, vehicles and ships can be effectively recorded. These spatiotemporal data contain a lot of potentially valuable patterns and knowledge. These potential models have very important research value in urban planning, national defense and military, location-based services, etc. In order to find out the moving target activity law, this paper uses co-occurrence mode analysis method to carry out co-occurrence analysis. Based on the analysis of the periodicity of co-occurrence mode, the main contents of this paper are as follows. In this paper, an algorithm of spatio-temporal cooccurrence pattern mining based on double-layer network is proposed. It is found from the space-time track data of moving objects of pedestrians, vehicles and ships that the spatio-temporal interest degree needs to be calculated, in order to simplify the calculation method of spatio-temporal interest degree. Based on the characteristics of spatio-temporal data, a two-layer spatio-temporal network modeling method is proposed in this paper. The spatio-temporal network effectively preserves the spatio-temporal relationship between spatio-temporal objects and space-time elements. The spatio-temporal frequency can be calculated quickly when calculating spatio-temporal interest. According to the element network layer, a large number of redundant candidate patterns can be reduced. In this paper, based on the space-time network, we propose an algorithm of cooccurrence pattern mining based on space-time network, which takes into account the effective period of elements. The weighted eigenvalue measurement of spatio-temporal frequency is proposed, and all spatio-temporal co-occurrence pattern sets are preserved by pattern chain. The experiment shows that when the same data set is used and the same results are obtained, This algorithm is more efficient than Celik algorithm and Wang algorithm. In this paper, an adaptive mining algorithm of moving target partial periodic co-occurrence pattern is proposed. The co-occurrence of three moving targets of vehicles and ships usually has partial periodicity. In this paper, the partial periodic pattern analysis is applied to the study of the co-occurrence law of moving targets. A partial periodic co-occurrence pattern adaptive mining algorithm is proposed, which adds the pattern validity degree according to the effective period of elements, and improves the calculation method of co-occurrence frequency. The adaptive algorithm is reflected in the determination of period span and confidence parameters. According to the time frame and initialization of spatiotemporal data, the method of adaptive determination of periodic span is given. According to the principle of maximum period span and the principle of preserving the periodic co-occurrence mode of all parts, the adaptive determination method of confidence parameters is given. Then, according to the property of periodic priori, the partial periodicity of long mode is determined first, then the partial periodicity of subset mode is considered, which reduces the calculation of co-occurrence frequency in co-occurrence analysis. The experiment shows that the algorithm proposed in this paper is compared with Apriori-like algorithm and Naive algorithm. The period span and confidence parameters can be calculated adaptively and accurately, and all of the periodic co-occurrence patterns are retained, which improves the running efficiency of mining patterns.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13

【参考文献】