Long-term trajectory classification and prediction of commercial vehicles for the application in advanced driver assistance systems

Current development of advanced driver assistance systems (ADAS), e.g., for collision mitigation are increasingly concerned about the protection of other road users. Environment perception provides objects like cars or pedestrians. A prediction of the system vehicle´s path is required to decide about the relevance of the objects for a system reaction or to reduce the CAN load in advance. A standard measure for the object criticality is the time-to-collision, where the system vehicle´s path is predicted under the assumption of constant acceleration and yaw rate or using lane markings. Lane markings often are not available on urban streets, and the vehicles do not necessarily follow the own lane, e.g., due to parked cars at the road side. This paper proposes an approach that uses maneuver classification based on a combination of the longest-common-subsequence method and a Bayesian classifier. The knowledge obtained about the maneuver in the classification step is used to predict the future trajectory in a parameterizable way. The approach is evaluated in comparison to a prediction with constant acceleration and constant yaw rate using recorded data from more than 20 hours of driving.