Search

In this report, we discuss about a data mining method for multi-modal in-vehicle sensor signals. Though several methods were previously proposed for symbolization of the multi-modal in-vehicle signals, the effectiveness of such method were not discussed in detail. We adopt the Bag-of-System (BoS) technique, a variation of the Bag-of-Features for sensor signals used in motion analysis, and evaluate the effectiveness for several complicated scenarios. According to the requirement of realistic practical data analysis, we extend the BoS technique and evaluate for scenarios such as lane change detection and turning estimation. According to the evaluation of the simple case, three behaviors at the intersections are successfully estimated. However, the technique exhibited lower performance for a complicated behavior such as lane change.

In this report, we have proposed a new data mining method to analyze vehicles' temporal behaviors. The Bag-of-Systems (BoS) representation was adopted to analyze the temporal behavior signals. The BoS representation was originally used for analyzing motion data from video image. We have extended the model of dynamical systems to analyze vehicles' temporal behaviors of both travel direction and rotational direction. By using this method, a symbol sequence was obtained for each trip. Therefore we could apply various data mining method to the obtained sequences. As for the first trial, a clustering method was applied to the sequences and characteristic segments to know the events happened on the vehicles were successfully found.