Event Detection Based On Location Information

With the advances in hardware technology of data acquisition, location information acquisition devices become more prevalent, resulting in a massive geographic data and spatial-temporal data. The research based on such data is an important filed of data mining, among which, the event detection is an area of focus for the researchers of spatial-temporal data mining. This paper presented the research of event detection based on location information, and the purpose was determining whether an event has occurred via statistical analysis of location feature information. The main idea of this paper is that the number of people that arrive a specific location at a specific time frame by taxi reflects human activity patterns, under normal circumstances it is relatively stable, if some abnormal event occurs that causes it significantly different from the number of people arriving at the same region at the same time frame of other different days by taxi, this incident can be detected by detecting those anomalies. This thesis consists of three parts: location feature information extraction; primitive event detection; complex event detection.In order to extract feature information of specific locations, the concept of region discretization was proposed, and through region discretization, GPS trajectories were converted to region trajectories. Finally, location feature information was extracted from region trajectories, laying the foundation of subsequent primitive event detection and complex event assembly detection. Additionally, the idea of region discretization and the processing method of GPS trajectory in this paper are also applicable to other research of trajectory data mining.This paper characterized the primitive event detection issue, pointed out the differences between the primitive event detection method proposed in this paper and other specific type event detection methods, analyzed the advantages and disadvantages of each other. The concept of common abnormality was proposed, based on which the definition as well as the formal description of the primitive event was given. Noise filtering methods of the primitive event detection results were discussed, and a specific primitive event noise filtering method was given. The mathematical principle of Gaussian mixture model based on of the EM algorithm was explained in this thesis, as well as the method of implementing clustering using Gaussian mixture model. Primitive event detection algorithm based on clustering was proposed, including the algorithm design and the algorithm flow chart.In this paper, the complex event detection was discussed. First of all, from the perspective of the relationship between primitive event and complex event, and how to detect complex event, complex event detection problem description as well as the definition of complex event was given. Secondly, as a focus of the complex event detection, spatial region adjacency was discussed, and the spatial region adjacency matrix was established. Then, the complex event detection algorithm was proposed, followed the details of algorithm design.Based on the extracted location features information, primitive event detection experiment was conducted, and filtered the noise of the experimental results, then, on the basis of the results of the primitive event detection, the complex event detection experiment was conducted, and the experimental results were analyzed and discussed. The validation of the experiment results shows that presented the primitive event detection algorithm and the complex event detection algorithm are accurate and reliable.