Although driving is a very complex process, many regularities can be identified. Humans very often repeat previously used routes, because driving along a known route (repeating driving pattern) could greatly reduce drivers' strain. It is believed that driving patterns could be easily extracted from signals generated by cheap and common navigation sensors. A system based on artificial intelligence could learn patterns and predict future driving events and thus provide support for the driver in the most typical navigation guidance tasks. The learning process provides adaptation of the system to a particular driver and his/her driving habits. Navigation support is based on previous driver experiences, enabling the system not only to detect possible problems with respect to the driving infrastructure, but also to detect differences in driver's behaviour over time, which could be caused by tiredness, influence of alcohol or a similar cause. In this paper results are presented in sub-symbolic short-term prediction of vehicle movements by neural networks and long-term prediction by Hidden Markov Models. (a) For the covering entry of this conference, please see ITRD abstract no. E202275.
Samenvatting