Reducing Uncertainty In Location Prediction Of Moving Objects In Road Networks

Consider a database which tracks moving objects in road network following a prespecified route. In a city environment, the number of moving objects can be large, and the frequency of updating objects’ location increases the load on the database management system. Hence, it is not feasible to update an object’s location constantly and explicitly. Instead, location of a moving object is stored as a dynamic attribute, e.g. motion vector function [3], whereby the location value is calculated when it is accessed, and it is updated when the parameters of the function change (speed, route, etc), and to increase location accuracy by comparing a possibly calculated location value with a measured value to keep the deviation bounded. Today, dead reckoning is used to implement the motion vector function, and to determine the current and future positions of an object based on knowledge of the underlying route network, the object’s pre-specified route and the object’s last position and speed [2]. However, tracking moving objects by dead reckoning techniques inherently introduces uncertainty in location determination and prediction since we do not really know where the object is actually located. Several cost factors are associated with determining the location of a moving object, i.e. the communication and update cost, uncertainty cost and deviation cost [3]. Communication and update cost is the cost involved in communicating data across the wireless network, and writing the data to the database. Deviation cost is associated with the actual linear distance by which the object’s location stored in the database and its actual location [3] deviates. Accounting for deviation in all possible directions introduces a region of uncertainty within which we locate the object and, thus, an uncertainty cost. Ideally, we would like to keep communication cost as well as deviation and uncertainty cost low; however, there is a trade-off between communication/update cost, and uncertainty cost. In this extended abstract, we propose an improved dead-reckoning policy that reduces the uncertainty in location prediction of moving objects in road networks while keeping update costs low. Location update policies as in [3] use linear deviation as the threshold parameter, e.g. an object deviates more than a certain threshold (e.g. 2 miles) from its assumed position on the assumed route. This technique relies on the availability of only one spatial sensor, i.e. a GPS receiver, and assumption about speed restrictions on the road network. We assume that we additionally have information about the direction in which an object is moving. Based on this information, we propose a hybrid dead reckoning policy that involves linear as well as an angular deviation as threshold parameters for the dead reckoning policy. This means we generate a location update when the linear deviation or angular deviation surpasses the defined threshold. In contrast to other approaches, we use the available information about angular deviation, and thus, can restrict the uncertainty significantly. If dl is the linear deviation threshold, w is the width of the road and da is the angular deviation threshold, then Uncertainty Cost without angular constraint: Ul = π * dl ^ 2 Uncertainty Cost with angular constraint for road networks Ua = 2 * dl * w if the actual angular deviation is less than da Reduction in Uncertainty Cost, Ru % = ((Ul – Ua) / Ul) * 100 Hence, Ru = (1 – (2 * w)/π * d)) * 100 Reducing uncertainty greatly influences a range query such as ‘retrieve all taxi’s that will be within 2 miles from airport in 15 minutes’, assuming 70% certainty. The target area is a circle of radius 2 miles with airport as center. The database predicts the future location of objects (15 minutes from now) as a point and the uncertainty as a region around the point (thus, a moving region), and checks whether at least 70% of the moving region falls within the target area. For a taxi, whose object region falling 70% within the target area, the uncertainty that it may not fall is 30% of the moving region. If we include an angular constraint, we can represent the object as a point, and its uncertainty area is a line if the angular deviation is zero. In this case, the remaining 30% uncertainty is of linear deviation, which leads to a more accurate decision-making. For future work in this area, we are interested in using the linear and angular deviation constraint at decision points in road networks to determine the location of moving object without knowledge of the pre-specified route. Hybrid location update policy also involves a tradeoff with communication cost since more data needs to be sent to the database, but as the communication cost gets cheaper faster, and network bandwidth increases as specified by International Telecommunications Union (ITU) [1], location tracking of moving objects can become more precise by employing the hybrid dead reckoning policy.