An unsupervised driving behavior pattern recognition algorithm based on clustering and LDA model

In order to achieve fully cooperative driving between the driver and vehicle, driving behavior pattern recognition is important. This paper presents a novel pattern recognition algorithm which can automatically translate the time series data of driving behavior captured with GPS-equipped mobile devices into annotated natural language. First, the time series segments of driving behavior are clustered into groups that represent driving senses in an unsupervised manner by using the adaptive K-means clustering. Then, the natural language expression of each driving sense is generated by using the Latent Dirichlet Allocation to associate the latent driving topics with each driving sense. Finally, automatic annotation of the driving behavior data can be achieved by calculating the predictive distribution of the annotations via estimated latent-driving topics. The proposed algorithm intuitively annotated 10093 pieces of driving behavior data, the driving topics achieved performance almost equivalent performance to human annotators.