Abstract
For the analysis of accidents and safety education, it is important that the risk evaluation for drive recorder data approximate human perception. This report deals with rear-end collision scenes in the near-miss incident database. It has a manual classification of risk levels, analyzed by using recorded event timing, acceleration, and recorded video. In this study, the deceleration behavior of the driver was inferred from the acceleration data. Also patterns were divided and characteristic timing also examined based on it. The risk evaluation method was examined for the preceding vehicle based on the change in the appearance of the preceding vehicle on the image 0.1 seconds before and after characteristic timing. In the proposed method, evaluation is performed by combining these methods. The validity was examined based on the relationship between the proposed method evaluated and the manual classification of risk levels. In the examination, we divided the cases into short or long times at sudden deceleration. As a result, when the deceleration time was short case, the evaluation value was tended to be higher than in the long case. But it was also seen that the evaluation value was similar when deceleration time was short and manual classification of low risk levels, and when it was long and high levels. By classifying the cases first based on the deceleration time, the proposed method has made it possible to deal with manual risk judgment in most cases.
