| 초록 |
In this paper, we propose a real-time implementation of the new Parking Lane Departure Warning System(PLDWS) algorithm for autonomous vehicle development. The proposed algorithm is for converting the top view image of the AVM camera into nine divided images and learning the deviation of parking line by studying the divided image based on deep learning. Wide angle cameras are mounted on the front, rear, left side, and right side of the real vehicle, and the top view image is obtained through the calibration of each camera. Considering that the visible range of the AVM varies according to the mounting position of the vehicle, the actual distance based on the AVM image is measured and designated as region of interest(ROI) to the front and rear, right and left 2-m regions(3 in the horizontal direction and 3 in the vertical direction). The proposed algorithm was implemented on NVIDIA Jetson TX2 board for deep learning implementation, and tested in Keras deep learning framework environment and Ubuntu 16.04 OS. It showed 98.01 % precision and 99.13 recall. The processing speed was 5.5 to 6 FPS. By using PLDWS, the driver can be notified with a warning about the parking situation through the camera, and the output of the PLDWS can be used for vehicle control when implementing the autonomous parking system. |
