基于运动特征的轨迹相似性度量研究

发布时间：2018-11-18 16:16

【摘要】：随着各种定位(如北斗、GPS等)和无线通讯技术的发展,人类获取了大量移动对象的时空轨迹数据,时空轨迹数据挖掘己成为当前的研究热点,在智能交通系统、气候监测、运动生态学等领域具有重要意义。运动特征是移动对象的运动参数如速度、加速度等表现出的可以区别于其他对象的征象或标志,它是移动对象的重要属性,能够反映移动对象的内在特点以及外部环境对其运动的影响。轨迹的相似性度量是轨迹数据挖掘的核心问题之一,本文研究基于运动特征的轨迹相似性度量,其可用于相似性查询及运动模式发现等应用中。本文以轨迹的运动特征为主线,对轨迹的相似性度量展开深入研究,改进并发展新的基于运动特征的轨迹相似性度量,并将其应用于相关应用中,论文主要工作和成果如下：1,在对现有运动参数和运动特征总结与凝炼的基础上,提出了基于层次运动特征和分类学习的轨迹相似性度量,并将其应用于移动目标识别。该相似性度量分别提取全局和局部运动特征而构成层次运动特征,全局特征利用高级的统计量来提取,局部特征从分割后的运动参数时间序列中提取,层次运动特征与分类学习方法支持向量机相结合构建相似性度量。在三个真实轨迹数据集上的实验表明,该相似性度量区分力强,与已有方法相比显著提高了移动目标的识别精度。2,提出了基于多重运动特征的轨迹相似性度量,并将其应用于基于多重运动特征的运动序列模式发现。该相似性度量借鉴数据立方体的思想,将多重运动参数时间序列进行量化和符号化表示,在多重运动特征值域空间中计算两符号间的距离,以此作为加权编辑距离的替换代价,最终以加权编辑距离作为相似性度量。该相似性度量反映了多重运动特征的演变趋势,即运动序列模式。将该相似性度量与谱聚类方法相结合进行运动序列模式发现,以大西洋飓风数据为例,通过气象文献中飓风的发生与运动规律验证了本方法的有效性,并分析了飓风多重运动特征的序列模式。3,提出了融合运动特征的轨迹时空相似性度量,并将其应用于基于运动特征的轨迹时空分布模式发现中。该相似性度量融合了空间距离、时间距离与运动特征距离。空间距离利用实数序列上的编辑距离(EDR, Edit Distance on Real Sequence)来度量,运动特征距离利用标准化的加权编辑距离(NWED,Normalized Weighted Edit Distance)来度量,时间距离利用两轨迹的起点距离、终点距离与轨迹的持续时间来度量,最终,通过加权平均方式将这三个距离有效、灵活地结合成融合运动特征的轨迹时空相似性度量。将该相似性度量与谱聚类方法相结合进行基于运动特征的时空分布模式发现,通过气象文献中飓风在空间与季节上的分布规律验证了本方法的有效性,并分析了基于飓风速度特征的轨迹时空分布规律。本文改进并发展了基于运动特征的轨迹相似性度量,推进了轨迹相似性的理论及应用研究,为轨迹数据挖掘提供了有价值的研究成果。
[Abstract]:With the development of various positioning (such as Beidou, GPS, etc.) and the wireless communication technology, the space-time track data of a large number of moving objects is acquired by human beings, and the time-space trajectory data mining has become the current research hotspot, and in the intelligent transportation system and the climate monitoring, and is of great significance in the fields of sports ecology and the like. The motion feature is the characteristic of the moving object, such as speed, acceleration, etc., which can be distinguished from other objects. It is the important attribute of the moving object, which can reflect the intrinsic characteristics of the moving object and the influence of the external environment on its movement. The similarity measure of the track is one of the core problems of the track data mining. This paper studies the track similarity measure based on the motion feature, which can be used in the application of similarity query and motion pattern discovery. In this paper, the motion characteristics of the track are the main line, the similarity measure of the track is further studied, the new track similarity measure based on the motion characteristics is improved and developed, and the method is applied to the related application, and the main work and the result of the paper are as follows: 1, On the basis of summarizing and refining existing motion parameters and motion characteristics, a track similarity measure based on hierarchical motion characteristics and classification learning is proposed and applied to mobile object recognition. the similarity measure is used for extracting global and local motion characteristics respectively to form a hierarchical motion feature, The hierarchical motion feature and the classification learning method support vector machine to construct the similarity measure. the experiments on the three real track data sets show that the similarity measure is strong, the recognition accuracy of the moving target is obviously improved compared with the existing method, and the track similarity measure based on the multiple motion characteristics is proposed, and is applied to a motion sequence pattern discovery based on multiple motion characteristics. the similarity measure refers to the idea of the data cube, and the multi-motion parameter time series is quantized and symbolized, the distance between the two symbols is calculated in the multi-motion characteristic value range space, Finally, a weighted edit distance is used as the similarity measure. The similarity measure reflects the evolution trend of the multiple motion characteristics, that is, the motion sequence mode. In this paper, the similarity measure is combined with the spectrum-based method to find that the Atlantic hurricane data is an example, the effectiveness of the method is verified by the occurrence and movement of the hurricane in the meteorological document, and the sequence pattern of the multi-motion characteristics of the hurricane is analyzed. In this paper, the space-time similarity measure of the motion feature is proposed, and it is applied to the trace-time distribution pattern discovery based on the motion feature. The similarity measure combines the spatial distance, the time distance and the moving feature distance. The spatial distance is measured by the edit distance (EDR, Edit Distance on Real Sequence) on the real sequence, the moving feature distance is measured using a normalized weighted edit distance (NWED, Normalised Weight Edit Distance), the time distance is measured using the starting distance, the end distance and the duration of the trajectory, and finally, the three distances are effectively and flexibly combined into a track space-time similarity measure of the fusion motion characteristic by a weighted average mode. By combining the similarity measure and the spectrum-based method, the spatial-temporal distribution pattern based on the motion characteristics is found, the effectiveness of the method is verified by the distribution law of the hurricane in the space and the season in the meteorological document, and the time-space distribution rule of the track based on the characteristic of the hurricane speed is analyzed. The paper improves and develops the track similarity measure based on the motion feature, and advances the theory and application of the track similarity, and provides valuable research results for the track data mining.
【学位授予单位】：南京师范大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：P208;P425

【相似文献】