Abstract
In order to promote researches aiming at reducing the number of traffic accidents, the near-miss incident database has been developed by summarizing near-miss incidents recorded by drive recorders mounted on taxis in Japan. The database is equipped with the distance measurement tool, by which researchers can obtain distance from the camera to an object in video images recorded by the drive recorders. However, it is a heavy task for researchers to obtain the distance data for a large number of incidents. Therefore, in this study, the authors proposes a method to obtain distance data to a target object automatically by using image processing. The algorithm performs tracking of the preceding vehicle by TLD, estimates the mounting position of the camera from the position of the fender mirrors and the bonnet, and calculate the vehicular gap using a perspective projection model. From the comparison with the vehicular gap obtained manually, it is indicated that the vehicular gap obtained by the proposed algorithm has satisfactory accuracy as the manual annotation process.
