An Effective Automatic Classification Method of Drive Recorder Data

Abstract

Drive recorders have been used more frequently in Japan on the transportation vehicle such as taxicabs, buses and trucks than before. The drive recorders can record the driving data including motion pictures in the traffic accidents and near-accidents, which are considered as “dangerous data”. Analysis and application of safety education using the “dangerous data” are useful to prevent traffic accidents. The drive recorder, however, can record a large number of “not dangerous data” which are not derived from the accident or near-accident data when the car passes the undulation or bumps on the road. The current useful method to extract the “dangerous data” exactly from all the raw data is the human observation and classification of the data while some automatic methods have been developed, which are not very exact to classify the data. The authors show a new method of automatic classification of the drive recorder data into more than two categories including “accidents”, “near-accidents” and “bound” etc., by using the waveforms of the acceleration data. The method is much more effective and reliable than the current automatic methods and it enables to build an automatic processing system for mass data from drive recorders to be classified exactly.