Identification of Aggressive Lane-changing Behaviour Based on Unsupervised Cluster Analysis
To effectively guide drivers to adopt safer lane-changing behaviours,this paper proposes a method to identify aggressive lane-changing behaviour based on a modified Self-Organising Mapping Neural Network(SOM-Kmeans)cluster analysis.Driving data and eye movement status are obtained by driving simulation equipment and eye movement equipment.Then,a change-point detection algorithm is applied to extract lane-changing behaviour event data from the multimodal dataset by combining the steering wheel angle and lateral gaze position.Afterwards,SOM-Kmeans cluster analysis is used to extract key feature parameters of driver lane changing behaviour and identify aggressive lane changing behaviour.The effectiveness of the SOM-Kmeans clustering method is compared with the density-based clustering algorithm(DBSCAN)and the fuzzy C-mean clustering algorithm(FCM),respectively,for the identification of aggressive lane changing behaviour.The results show that SOM-Kmeans is able to classify aggressive lane-changing behaviour into two types:emergency lane-changing and squeezing lane-changing.The proposed method can establish the corresponding behavioural indicators and thresholds,and identify the lane changing behaviour as aggressive when the acceleration fluctuation in the process of lane changing is greater than 8.22 m·s-3 and the steering wheel angle is greater than 0.83(°)·s-1.Based on aggressive lane changing behaviour,when the lane changing gap is less than 7.5 m and the duration of the lane changing is greater than 10.3 s,the lane changing is identified as crowded lane changing,otherwise it is emergency lane changing behaviour.Crowded lane changing behaviours are mostly found in mandatory lane changing with heavy congestions,and emergency lane changing behaviours are mostly found in free lane changing with low-to-moderate traffic densities.The accuracy of the proposed method identification is 92.5%when compared with the traditional cluster analysis.The proposed method can identify the types of aggressive lane-changing behaviour in a more detailed way,and the results of the study can be used as a way to assess whether there is a deviation from the normal lane changing behaviours of a driver and to measure the driver's lane changing habits.The results of the two-layer clustering can also be used as a referential criteria of the radical type of lane changing behaviours.
NETL
NSTL
万方数据
维普
