ドライブレコーダによる交通流データ取得システムの開発

抄録

Recently, traffic congestion due to increase of the number of vehicle possession and freight distribution has been become a serious social problem. In order to solve this problem, various examinations using traffic simulation systems have been carried out before now. However, the current simulation system does not have sufficient reliability and reproducibility on the traffic behavior, such as car appearances, driving lane change and running speed of cars etc., because those are defined by random numbers and arbitrary conditions. From these reasons, traffic flow measurement system needs to be developed to improve reliability of the system and reproducibility of real traffic on the simulation system. In our previous work, the mAP value of classifier using R-CNN was 61.7 and the recognition rate of identical vehicle did not reach satisfactory level. In this study, we will build a classifier with the better accuracy and implement SURF feature method to detect identical objects. Furthermore, we add a new function that objects are mapped to Google Earth to be able to identify the position and the appearance time of the vehicle at a glance. As a result, these approaches are sufficiently successful on grasping traffic condition around our own car being equipped with a drive recorder. In the future, those are necessary to examine the features that recognize objects and to reduce the image processing time.