Prediction of crossroad passing using artificial neural networks

One of the key issues while minimizing the cost of a continuous tracking of moving object is precision of predicted trajectory. Analysis of location data showed that car drivers often participate in a typical spatio-temporal motion patterns. Precision of location prediction and thus client-server tracking communications can be improved by learning such motion patterns. Paper proposes a novel approach to make predictions about where to the driver will turn at the crossroad as these are the points where trajectory can be branched in the road network environment. Approach applies artificial neural networks(ANN) which are used to learn temporal patterns of crossroad passing and predict transitions through the crossroad. Several neural networks' structures are presented and analyzed, also ANN based method is compared to statistical techniques. Proof of concept of the proposed prediction method was analyzed using real and artificial data sets. The potential of the crossroad passing prediction is discussed.