文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > 工程科技 > 机械/仪表 > A New Trajectory Clustering Based on Paritition-Cluster-Extration
侵权投诉

A New Trajectory Clustering Based on Paritition-Cluster-Extration

2013 International Conference on Computational and Information Sciences

A New Trajectory Clustering Based on Paritition-Cluster-Extration

Jiashun Chen

Dechang Pi

College of Computer Science and Technology

NUAA Nanjing, China

E-mail: nuaacs@yahoo.com.cn

College of Computer Science and Technology

HHIT Lianyungang, China E-mail:1976139@163.com

Abstract— In this paper, we proposed a new trajectory behavior characteristics are analyzed to identify clustering based on partition-cluster-extract (PCE). Firstly, homogeneous events [5]. The main fields of this paper are some relative definitions are defined, and then we proposed in surveillance system trajectory, which are different from a new partition method named PCE, which is based on our topics.As a result of mixture regress model was based method of partition-group framework. Finally, on the basis

on independence, with increase of trajectory number,

of PCE and relative definitions, we proposed a new lagorithm named partition-cluster-extract trajectory mixture model may be invalid. Ng et al [6] proposed a

mixture of mixed effects model to capture the correlation clustering (PCETC). Some experiments on different datsets

by using random effects. New model can be adjusted show that the new algorithm in this paper is more efficient

according to individual probabilities of membership. than TRACLUS, at the same time, the algorithm has nearly

the same cluster quality as TRACLUS. Aiming at spare data, literature [7] proposed a new

method named ACENT (Attribute-aided Classification of

Key words— cluster; partition-cluster-extract; trajectory Entangled Trajectories) to reconstruct spatiotemporal

trajectories. According to input parameters such as spatiotemporal coordinates and some attributes, ACENT I. INTRODUCTION

can classify and cluster trajectory data. Reconstructing

Trajectory clustering is an interesting branch of

trajectory of this method is to linked some trajectory

clustering and many new trajectories clustering algorithm

segments to form longer trajectory and then to cluster.

are coming up. Pao et al [1] analyzed different types

This method is contrary to our idea. We partition a long

trajectories including mouse trace, handwritten characters

trajectory into several parts to find every segments feature,

and animals trajectory and proposed a general user

and then according to this feature, clustering this

verification approach based on these trajectories. They

trajectory segments [8]. Aiming at variables measured

describe trajectory behavior by using Markov chain model

repeatedly, Literature [9] proposed a new idea about the

with Gaussian distribution. Abraham et al[2] deal with

joint evolution of several variables and find that studying

the variations in applying a spatiotemporal similarity

the joint is more meaningful. Literature [9] and [10]

measure with given Points of Interest (POI) and Time of

applied k-means algorithm to find joint trajectories and

Interest (TOI),treating spatial similarity as a combination

cluster. This method manages missing value, and also

of structural and sequence similarities that is evaluated

provides some quality criteria to find better partition.

using the techniques of dynamic programming. The

K-means algorithm is applied in point data set, if it is used

similarity set thus formed will be used by the remote

in trajectory data, clustering resutls are not obvious and

database to broadcast trigger-based messages to

that is not good idea. To find joint and partition is a

participating vehicles in a neighborhood for future route-

difficult thing. Literature [11] applied FCM (fuzzy

and information-sharing activities.Algorithm in [3]

c-means) in typhoon movement trajectory and compare

extracted trajectory from image sequence based on cluster

with hierarchical clustering, normalized cut techniques

model. This regression mixture model for cluster

and k-means algorithm. Experiments show that FCM has

trajectories of variable length is a sparse, translation

good performance. FCM has advantage, however, most

invariant. Parameters in this model are selected by using

trajectories are long segments, and different parts include

Expectation-Maximization algorithm. Experiments

various information. Therefore, a long trajectory is not

demonstrated this algorithm is robust and more accurate.

partitioned and clustering can not get better results.

A trajectory clustering algorithm for video surveillance

In order to solve the problem that how to partition

system is proposed by Piciarelli [4]. They organized

trajectory and find similar parts, we proposed a new

trajectory clustering on-line by using tree-like structure, at

partition method based on TRACLUS. This method

the same time, some probability information including

includes three steps: partition trajectory, cluster and

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email