Characteristics analysis on speed time series with empirical mode decomposition as vehicle driving towards an intersection

Abstract

In this paper, we explore the characteristics of vehicle speed time series which described the processes that a driver finishing a specific driving task with different driving operations. Three types of vehicle driving behavior like driving towards an intersection for turn-left, driving for turn-right, and driving for go-straight are designed as a set of real vehicle driving experiments to be carried out on an urban road. Similar to the expected, the collected speed time series of all driving behavior types tend to be non-linear and non-stationary. Therefore, empirical mode decomposition (EMD) is introduced to analyze the characteristic values of speed time series intrinsic mode functions (IMF) and residues. After decomposing, there are 4 levels of IMF with a residue exist existed in the speed time series of turn-left driving behavior, as well as 3 levels in turn-right and 5 levels in go-straight. All the first level IMF of three types of vehicle driving behavior have relatively high frequencies which could be regarded as systematic errors of vehicle speed sensors. As the decomposition continued, subsequent IMF frequencies become lower but average amplitudes have different change trends which could help identifying the driving behavior types. All residue curves are firstly monotone increasing and then monotone decreasing, but the occurrence time of residue maximums are inconsistent. Through this research, we can distinguish the driving behavior type between turn-left, turn-right and go-straight with those vehicle driving behavior time series characteristic values and their changing trend. If all those judgment and statistics of characteristic values be implemented by vehicular industrial control computes, it would improve driving behavior recognition or prediction performances of an advanced driving assistance embedded on vehicle.