Finding long and similar parts of trajectories

Buchin, Kevin; Buchin, Maike; Kreveld, Marc van; Luo, Jun

doi:10.1145/1653771.1653813

Cited by 27 publications

(29 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, it would be interesting to look at other types of clustering methods with these alignments, for instance density based ones. Finally, we would like to look at similarity measures from computational geometry, as for instance in [3,15].…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Clustering Vessel Trajectories with Alignment Kernels under Trajectory Compression

Vries

Someren

2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

In this paper we apply a selection of alignment measures, such as dynamic time warping and edit distance, to the problem of clustering vessel trajectories. Vessel trajectories are an example of moving object trajectories, which have recently become an important research topic. The alignment measures are defined as kernels and are used in the kernel k-means clustering algorithm. We investigate the performance of these alignment kernels in combination with a trajectory compression method. Experiments on a gold standard dataset indicate that compression has a positive effect on clustering performance for a number of alignment measures. Also, soft-max kernels, based on summing all alignments, perform worse than classic kernels, based on taking the score of the best alignment.

show abstract

Section: Discussionmentioning

confidence: 99%

“…The w = 1 setting means that the average difference between two points in the space dimension is roughly equal to the average difference in the time dimension. 3 To investigate the influence of trajectory compression on the clustering performance we used 4 compression settings. The first setting being = 0, thus, we apply no compression.…”

Section: Methodsmentioning

confidence: 99%

Clustering Vessel Trajectories with Alignment Kernels under Trajectory Compression

Vries

Someren

2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…Our new lower bound filtering approach has ( 1 Se + 1 − α) • |T | calls. Our lower bound filtering approach is better than our previous columnwise lookup table approach if 1 Se + 1 − α < 1 (i.e., α > 1…”

Section: Comparison Of Cost Models Of the Three Ap-mentioning

confidence: 92%

“…Se ), and is worse otherwise. In practice, in order to make α > 1 Se , we need to select a preferable S e , k, and track envelope formation strategy. According to our discussion on factors controlling α in Section 4.2.3, in order to have high pruning ratio α, we should prefer a smaller k and S e , and track envelope formation based on "Jaccard coefficient", or "overlap" rather than "random".…”

Section: Comparison Of Cost Models Of the Three Ap-mentioning

confidence: 99%

“…Related work : Figure 2 summarizes related trajectory mining techniques in the existing literature, which can be categorized into two groups, geometry space based and network space based. Techniques in geometry space use Euclidean distances [23,1,16], and thus cannot model network travel costs (e.g., roads along two sides of a river are close in geometry space but distant in network space). Techniques in network space include "hot" or popular route approaches, such as densitybased [17,14,4], frequency-based [15] and hierarchical clustering [26,22].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Identifying K Primary Corridors from urban bicycle GPS trajectories on a road network

Jiang

Evans

Oliver

et al. 2016

Information Systems

View full text Add to dashboard Cite

Given a set of GPS tracks on a road network and a number k, the K-Primary-Corridor (KPC) problem aims to identify k tracks as primary corridors such that the overall distance from all tracks to their closest primary corridors is minimized. The KPC problem is important to domains such as transportation services interested in finding primary corridors for public transportation or greener travel (e.g., bicycling) by leveraging emerging GPS trajectory datasets. However, the problem is challenging due to the large amount of shortest path distance computations across tracks. Related trajectory mining approaches, e.g., density or frequency based hot-routes, focus on anomaly detection rather than identifying representative corridors minimizing total distances from other tracks, and thus may not be effective for the KPC problem. Our recent work proposed a k-Primary Corridor algorithm that precomputes a column-wise lookup table of network Hausdorff distances. This paper extends our recent work with a new computational algorithm based on lower bound filtering. We design lower bounds of network Hausdorff distances based on the concept of track envelopes and propose three different track envelope formation strategies based on random selection, overlap, and Jaccard coefficient respectively. Theoretical analysis on proof of correctness as well as computational cost models are provided. Extensive experiments and case studies show that our new algorithm with lower bound filtering significantly reduces the computational time of our previous algorithm, and can help effectively determine primary bicycle corridors.

show abstract

Clustering Spatio-temporal Trajectories Based on Kernel Density Estimation

Zhang

Deng

Weghe

2014

Computational Science and Its Applications – ICCSA 2014

View full text Add to dashboard Cite

Abstract. The data mining from spatio-temporal trajectories of moving objects has been paid much attention since the last decade and has been considered as one of the new research fields that are attracting great interest. As is known, clustering analysis is one of the most effective tools that are commonly used in data mining. Based on this, in this article, a kernel density estimation based approach is proposed towards the clustering of spatio-temporal trajectories, with the aim to investigate the spatio-temporal clustering of trajectories. In this approach, firstly, the spatio-temporal neighborhood of each trajectory unit is built. Secondly, the trajectory unit sets that are with high densities are extracted in terms of the number of trajectory units their neighborhoods contain. Thirdly, the spatio-temporal kernel density of each trajectory unit is calculated with the Gauss kernel function. What follows next is to search all the density-attracting lines from the extracted trajectory unit sets. Finally, the spatio-temporal clustering of trajectories is executed based on the density-attracting lines, each of which is regarded as the center of a cluster. Last but not least, the feasibility and efficiency of the approach is validated using a real trajectory dataset.

show abstract

Finding long and similar parts of trajectories

Cited by 27 publications

References 37 publications

Clustering Vessel Trajectories with Alignment Kernels under Trajectory Compression

Clustering Vessel Trajectories with Alignment Kernels under Trajectory Compression

Identifying K Primary Corridors from urban bicycle GPS trajectories on a road network

Clustering Spatio-temporal Trajectories Based on Kernel Density Estimation

Contact Info

Product

Resources

About