多模式移动对象不确定性轨迹预测模型

A Multiple-motion-pattern Trajectory Prediction Model for Uncertain Moving Objects

以移动设备、车辆、飞机、飓风等移动对象不确定性轨迹预测问题为背景,将大规模移动对象数据作为研究对象,以频繁轨迹模式挖掘、高斯混合回归技术为主要研究手段,提出多模式移动对象轨迹预测模型,关键技术包括:1)针对单一运动模式,提出一种基于频繁轨迹模式树FTP-tree的轨迹预测方法,利用基于密度的热点区域挖掘算法将轨迹点划分成不同的聚簇,构建轨迹频繁模式树,挖掘频繁轨迹模式预测移动对象连续运动位置.不同数据集上实验结果表明基于FTP-tree的轨迹预测算法在保证时间效率的前提下预测准确性明显优于已有预测算法.2)针对复杂多模式运动行为,利用高斯混合回归方法建模,计算不同运动模式的概率分布,将轨迹数据划分为不同分量,利用高斯过程回归预测移动对象最可能运动轨迹.实验证明,相比于基于隐马尔科夫模型和卡尔曼滤波的预测方法,所提方法具有较高的预测准确性和较低的时间代价.

This study aims to solve the problem of predicting uncertain trajectories of moving objects, including mobile devices, vehicles, airplanes, and hurricanes. In order to design a general schema of trajectory prediction on large-scale moving objects data, techniques of frequent trajectory patterns mining and Gaussian mixture regression model are employed,and a multiple-motion-pattern trajectory prediction model is proposed. The proposed key techniques include： 1） as for simple motion patterns, a new trajectory prediction algorithm based on frequent trajectory pattern tree（FTP-tree） is proposed, which employs a density based region-of-interest discovery approach to partition a large number of trajectory points into distinct clusters. Then, it generates a frequent trajectory pattern tree to forecast continuous locations of moving objects. Experimental results show that the FTP-tree based trajectory prediction algorithm performs better than existing prediction approaches with the guarantee of time efficiency. 2） Gaussian mixture regression approach is used to model complex multiple motion patterns, which calculates the probability distribution of different types of motion patterns, as well as partitions trajectory data into distinct components, in order to predict the most possible trajectories of moving objects via Gaussian process regression. Experimental results show a high accuracy and low time consumption on trajectory prediction, as compared to the hidden Markov model approach and the Kalman filter one.