Driving Behavior Primitives Clustering Based on Temporal Characteristics

The semantic analysis of driving behaviors is particularly important in autonomous driving. In order to better understand driving behaviors, a temporal characteristic based Latent Dirichlet Allocation (t-LDA) is proposed to analyze different driving behavior primitives. In this algorithm, sliding windows are used to divide driving data into window data, and the continuous window data are converted into discrete driving words through k-means. Based on these driving words, driving behavior primitives are divided into kinds of categories with semantics. Finally, superiorities of the proposed t-LDA are verified by comparing it with the traditional LDA. The results are instructive to get more accurate descriptions of driving behaviors.