Abstract
Near-miss database consists of video data of real-world dangerous situations. It can be classified into the overt dangerous situation, which is obviously approaching accident, the potentially dangerous situation which has probability to get into the overt dangerous situation, and the normal driving situation. In order to assure safety driving of autonomous vehicle, it is necessary to evaluate whether the vehicle would be operated according to the scenarios which avoid not only the overt dangerous situations but also the potentially dangerous situations. In this paper, using technologies of object detection and measurement on the near-miss video data, and using theoretical concept of potential risk indicator, rear-end type near-miss data were analyzed, the potential risk indicators were calculated, and the potentially dangerous situations were defined. Then, a novel evaluation method of safer autonomous driving vehicles is proposed.
