In the context of the European project SimuSafe, various studies using the Honda Riding Trainer (HRT) simulator have been carried out. The resulting simulations from these studies are designed on the basis of a set of predefined scenarios and aim to analyze the motorcyclists’ behavior when interacting with the infrastructure. Furthermore, the analysis of the risk incurred by the riders’ maneuvers is of utmost importance to ensure their safety in realistic situations. However, the nature of road patterns (left or right turns, roundabouts and straight lines) is unknown in advance and should be deduced from the rider’s behavior or by extracting some infrastructure-based features from GPS track. This paper concentrates on the segmentation of trajectories and identification of different road patterns using sensory data and extracted features with the aim to facilitate the analysis of potential risks related to each pattern. To conduct this analysis, three non-supervised machine learning techniques are evaluated and their performances are compared. Finally, an exploratory cross analysis between the identified situations and rider’s dynamic behaviors allows for a more in-depth understanding of riders’ decisions.
展开▼