Driving safety in the mixed traffic state of autonomous vehicles and conventional
vehicles has always been an important research topic, especially on highways
where autonomous driving technology is being more widely adopted. The merging
scenario at highway ramps poses high risks with frequent vehicle conflicts,
often stemming from misperceived intentions [1].
This study focuses on autonomous and conventional vehicles in merging scenarios,
where timely recognition of lane-changing intentions can enhance merging
efficiency and reduce accidents. First, trajectory data of merging vehicles and
their conflicting vehicles were extracted from the NGSIM open-source database in
the I-80 section. The segmented cubic polynomial interpolation method and
Savitzky–Golay filtering are utilized for data outlier removal and noise
reduction. Second, the processed trajectory data were used as input to a hybrid
Gaussian hidden Markov (GMM-HMM) model for driving intention classification,
specifically lane-change collision-avoidance and lane keeping. The K-means
algorithm is used to initialize the model parameters, and the
expectation–maximization (EM) algorithm is employed for parameter iteration.
Finally, through validation on the testing set, the mixed Gaussian hidden Markov
model achieves a lane-change intention recognition accuracy of over 95% for
conflicting vehicles and outperforms the support vector machines (SVM) model and
the long–short-term memory (LSTM) network. It can be applied to the humanized
design of intelligent vehicle lane-change strategies, effectively reducing
lane-change risks and improving driving safety.
