Real Time Detection and Tracking of Spatial Event Clusters
We demonstrate a system of tools for real-time detection of significant clusters of spatial events and observing their evolution. The tools include an incremental stream clustering algorithm, interactive techniques for controlling its operation, a dynamic
- PDF / 1,107,372 Bytes
- 4 Pages / 439.37 x 666.14 pts Page_size
- 73 Downloads / 225 Views
Abstract. We demonstrate a system of tools for real-time detection of significant clusters of spatial events and observing their evolution. The tools include an incremental stream clustering algorithm, interactive techniques for controlling its operation, a dynamic map display showing the current situation, and displays for investigating the cluster evolution (time line and space-time cube).
1
Problem Setting
Spatial events are physical or abstract entities with limited existence times and particular locations in space, for example, lightning strikes or mobile phone calls. A spatial event is characterized by its start and end times (which may coincide), spatial coordinates, and, possibly, some thematic attributes. We assume that occurrences of spatial events are registered, e.g., by sensors, and corresponding data records are immediately sent to a server. The resulting data stream needs to be monitored. We consider monitoring scenarios in which each individual event is not significant whereas spatio-temporal event clusters (i.e., occurrence of multiple events closely in space and time) may require observer’s attention. For example, moving vehicles may emit low speed events when their speed drops below a certain threshold. It is neither feasible nor meaningful to attend to every such event, but a spatio-temporal cluster of low speed events sent by several cars may deserve observer’s attention as a possible indication of a traffic jam. After detecting a cluster, the observer may need to trace its further evolution, i.e., changes in the number of events, number of vehicles involved, spatial location, shape, and extent. The task is to support the observer in detecting the emergence and tracking the evolution of spatial event clusters in real time. © Springer International Publishing Switzerland 2015 A. Bifet et al. (Eds.): ECML PKDD 2015, Part III, LNAI 9286, pp. 316–319, 2015. DOI: 10.1007/978-3-319-23461-8_38
Real Time Detection and Tracking of Spatial Event Clusters
2
317
Approach
We apply clustering techniques to separate spatio-temporal event concentrations (clusters) from scattered events (noise). In fact, the problem setting requires an analog of a density-based clustering method capable to process a data stream in real time. However, the existing stream clustering methods are oriented to somewhat different problem settings. The main problem they address is the memory limitation. Assuming that all data cannot fit in the memory, the methods summarize incoming data on the fly and keep only the summaries (micro-clusters) but not the original data items. Many streaming algorithms assume a two-phase approach: micro-clusters are created and maintained during an online phase and post-processed (e.g., merged into larger clusters) during an offline phase. This general framework is instantiated with different approaches to creating micro-clusters. The main representatives are CluStream [1] and DenStream [2] doing partition-based and density-based clustering, respectively. CluStream partitions an initial portion of
Data Loading...
