Interval-Orientation Patterns in Spatio-temporal Databases

In this paper, we present a framework to discover a spatio-temporal relationship patterns. In contrast to previous work in this area, features are modeled as durative rather than instantaneous. Our method takes into account feature’s duration to capture the temporal influence of a feature on other features in spatial neighborhood. We have developed an algorithm to discover a temporal-spatial feature interaction patterns, called the Interval-Orientation Patterns. Interval- Orientation pattern is a frequent sequence of features with annotation of temporal and directional relationships between every pairs of features. The proposed algorithm employs Hash-based joining technique to improve the efficiency. We also extend our approach to accommodate an incremental mining as updates in real world spatio-temporal databases are common. The incremental algorithm employs an optimization that is based on previously generated patterns to prune the non-promising candidates early. We evaluate our algorithms on synthetic dataset to demonstrate its efficiency and scalability. We also present the patterns identified from real world drought, vegetation and video action databases. We also show that the patterns discovered from video dataset can improve the classification accuracy of activity recognition.