Efficient Detection of Emergency Event from Moving Object Data Streams

The advance of positioning technology enables us to online collect moving object data streams for many applications. One of the most significant applications is to detect emergency event through observed abnormal behavior of objects for disaster prediction. However, the continuously generated moving object data streams are often accumulated to a massive dataset in a few seconds and thus challenge existing data analysis techniques. In this paper, we model a process of emergency event forming as a process of rolling a snowball, that is, we compare a size-rapidly-changed (e. g., increased or decreased) group of moving objects to a snowball. Thus, the problem of emergency event detection can be resolved by snowball discovery. Then, we provide two algorithms to find snowballs: a clustering-and-scanning algorithm with the time complexity of O(n(2)) and an efficient adjacency-list-based algorithm with the time complexity of O(nlogn). The second method adopts adjacency lists to optimize efficiency. Experiments on both real-world dataset and large synthetic datasets demonstrate the effectiveness, precision and efficiency of our algorithms.