Modeling for Cluster-Based Correlation of Safety Driving Events with Time and Location

Advanced driver assistance systems (ADAS) are general systems developed to enhance safety driving of an individual vehicle. In this dissertation, a type of ADAS, named safety driving assistance system, is proposed to lower the potential driving risk caused by severe driving events to enhance the safety driving of an individual vehicle. The safety driving assistance system identifies the severe driving events and the occurrence of the events to infrastructure, notifies the events temporal and spatial concentration and variation, and models the concentration and variation in form of the event count data time series modeling to evaluate and predict the potential driving risk. Safety driving assistance system uses a designed intelligent analyzer which is a systematic procedure to identify severe driving events occurrence correlation with time and location. The proposed procedure, which is constructed based on batch clustering and real-time clustering techniques, incorporates historical and real-time data to recognize the time and location of severe driving events and simulate the variation of severe driving events distribution and concentration with respective to time and location, respectively. Batch clustering is implemented with the combination of subtractive clustering and fuzzy c-means clustering to generate clusters representing the initial correlation patterns. Real-time clustering is then developed to create and update real-time correlation patterns on the foundation of the batch clustering using evolving Gustafson-Kessel Like (eGKL) algorithm. Historical and real-time data of operating vehicles acquired from data acquisition and wireless communication platform (DAP), constructed by Ford Motor company, are used to validate the proposed strategy. Batch clustering reveals the severe driving events distribution and concentration in geographical domain at different time. Real-time clustering provides and updates the variation of the intra-correlation and inter-correlation of different regions. Driver can be notified of the potential severe driving locations through maps showing the driving routes. Through the variation of the correlation, drivers can recognize the events occurrence at different time and location. The variation of the correlation can be presented by events count data time series. Four models are proposed to describe time series of event count data in a region and to predict the future event count in the region. ARIMA and STARIMA modeling procedures account more on the aspect of the time series autocorreation in temporal domain and spatial domain. Generalized linear model (GLM) with Poisson distribution accounts more on the aspect of the natural distribution property of severe driving event. Hidden Markov Model (HMM) is attempted to describe and predict the event count data in a deep reasoning that the stochastic process of severe driving event occurrence in different regions is generated from different Poisson distribution components following certain transition logic. The four models are all validated by actual data and demonstrated their adequacy.