传感器网络基于轨迹聚类的多目标跟踪算法

Multiple Objects Tracking Algorithm by Trajectories Clustering in Sensor Network

本文重点研究传感器网络中能源高效的多目标跟踪问题.根据轨迹相似性对跟踪目标聚类,利用组对象跟踪实现所有对象的跟踪,能够有效地减少传输能耗,延长网络寿命.由于测量误差、低采样率以及环境干扰,很难获取目标的精确位置,因此轨迹数据存在固有的不确定性.忽略这种不确定性会降低轨迹挖掘质量,从而影响目标跟踪.提出基于不确定性轨迹挖掘的组对象跟踪方法.轨迹挖掘阶段首先为所有跟踪目标建立马尔科夫链模型,然后给出一种新的不确定轨迹相似性的度量,最后给出不确定轨迹聚类算法UTK-means对目标分组.组对象跟踪阶段向基站周期性地更新组中心轨迹的位置.实验结果验证了本文方法具有较高的聚类质量和节能效率.

This paper studies the problem of energy efficiently multiple objects tracking in wireless sensor network.Tracking a group of similar objects by trajectories clustering instead of tracking every single object can reduce the transmission energy and prolong network lifetime.However,due to hardware limitations of sensors,lowsampling rate,as well as complex natures of surroundings,the locations of moving objects may not be precisely obtained,thus uncertainty inherently exists in trajectories.Ignoring the uncertainty will reduce the accuracy of the mining algorithms and affect the object tracking.This paper presents group objects tracking method based on uncertain trajectories clustering.In trajectory mining phase,we build Markov chain model for each uncertain trajectory and then give a newtrajectory similarity measure.Finally,we present the uncertain trajectory clustering algorithm UTK-means to cluster the similar trajectories into groups.In group tracking phase,the locations of groups are transmitted to Base Station periodically.Experiments results demonstrate the good mining quality and energy-saving efficiency of our method.