깊이정보를 이용한 영상주차유도시스템 카메라 자동 보정

초록

Recently, video parking guidance systems have been becoming increasingly popular for the efficient management of parking spaces. Although they are simply used as a guide for finding a place to park, they are also linked to a parked vehicle search service to find a place where a vehicle driver is parked by using recognition technology. As parking occupation recognition and license plate recognition occur through camera input, it is important to maintain the quality of original camera input over time. However, in the case of parking lots, vibration occurs frequently and the equipment may become distorted over time, while the initial setting value of the camera may change due to physical impulse; thus, continuous performance maintenance is required through camera auto-calibration. Therefore, in this paper, Monodepth is used for automatic correction to extract depth information from a single image, which enables the maintenance of not only a certain distance between camera and object but also the accuracy of both object and number recognition by adjusting the camera settings. Hence, it was possible to maintain initial accuracy while automating maintenance.

키워드