Data-driven situation awareness algorithm for vehicle lane change

A good level of situation awareness is critical for vehicle lane change decision making. In this paper, a Data-Driven Situation Awareness (DDSA) algorithm is proposed for vehicle environment perception and projection using machine learning algorithms in conjunction with the concept of multiple models. Firstly, unsupervised learning (i.e., Fuzzy C-Mean Clustering (FCM)) is drawn to categorize the drivers' states into different clusters using three key features (i.e., velocity, relative velocity and distance) extracted from Intelligent Driver Model (IDM). Statistical analysis is conducted on each cluster to derive the acceleration distribution, resulting in different driving models. Secondly, supervised learning classification technique (i.e., Fuzzy k-NN) is applied to obtain the model/cluster of a given driving scenario. Using the derived model with the associated acceleration distribution, Kalman filter/prediction is applied to obtain vehicle states and their projection. The publicly available NGSIM dataset is used to validate the proposed DDSA algorithm. Experimental results show that the proposed DDSA algorithm obtains better filtering and projection accuracy in comparison with the Kalman filter without clustering.