Heuristic Methods for Automating Event Detection on Sensor Data in Near Real-Time
Abstract
Moving target indicator (MTI) analysts in the field are responsible for processing the increasing amounts of live streaming data. Analysts manually access unique data sources through a set of tools, and perform analysis on the available data. Operationally, analysts can only concentrate on small areas of interest and are subject to attentional blindness. Abnormalities in the periphery are often not detected until the forensic stage. Analysts are in need of assistance in performing data analysis. This paper presents the implementation of a heuristic-based stream mining approach for cueing the analyst user on geospatial temporal patterns (termed "event" for this effort) in near real-time. This approach is designed to aid analysts in detecting noteworthy events scattered within the overabundance of data, a problem which is well-documented and recognized. The implementation involves two phases: the isolation of areas of unusual activity using density grids, followed by event detection within those areas. Four analyst-identified events - starburst, inverse starburst, fanning, and inverse fanning - were identified for automated detection using these techniques. The event detection method was employed as a service within the Sensor Data and Analysis Framework (SDAF). The algorithm implementation and evaluation produced findings and informal user feedback. The results of this effort aids in establishing the foundation for near real-time event detection in MTI data analysis.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2010
- Accession Number
- AD1108458
Entities
People
- Barry Lai
- Daniel Mauer
- Eddy Cheung
- Jennifer Casper
- Jing Hu
- Peter Leveille
- Ronald Albuquerque
Organizations
- MITRE Corporation