Probabilistic Model To Identify Movement Patterns In Geospatial Data
Abstract
The task of trying to determine the movement pattern of objects based on available databases is a daunting one. Tracking the movement of these dynamic objects is important in different areas to understand the higher order patterns of movement that carry special meaning for a target application. However this is still a largely unsolved problem and recent work has focused on the relationships of moving point objects with stationary objects or landmarks on a map. Global Position System (GPS) is a widely used satellite-based navigation system. Popular use of these devices has produced large collections of data, some of which have been archived. These archived data sets and sometimes real time GPS data are now readily available over the internet and their analysis through computational methods can generate meaningful insights. These insights when applied appropriately can be used in everyday life. The purpose of this research is to make the case that automated analysis can provide insight that can otherwise be difficult to achieve due to the large volume and noisy characteristics of GPS data. We present experiments that have been performed on one of these archived databases which contain GPS traces of 536 yellow cabs in the San Francisco Bay area. Using data analysis, we determine the most visited tourist destinations within the San Francisco Bay area during the time period of the captured data. We also propose a probabilistic framework, which determines the probability of a new routing pattern using previous patterns. We use simulated routing patterns built on the same data format as that of the San Francisco cab data to predict the possible routes to be taken by a vehicle. All the probability calculations performed are done using Bayes' theorem of conditional probability formula.