The Real-time Implementation for the Parking Line Departure Warning System

Abstract

The PLDWS (Parking Line Departure Warning System) is one of the important ADAS (Advanced Driving Assistance System) functions when a driver recognizes the parking situations. We proposed the new real-time PLDWS algorithm and its real-time implementation based on the NVIDIA Jetson TX2, which is an embedded board environment. We showed the processing speed that is applicable to actual production vehicle and the same performance result as PC environment. A wide-angle camera is mounted on the front, rear, left and right sides of the vehicle, AVM (Around View Monitoring) images are obtained by synthesizing one Top-View image using four camera input images that have been calibrated. Considering that the visible range of the AVM system varies depending on the mounting position of the vehicle, the actual distance of the AVM image is measured and designated as ROI (Region of Interest) to the front and rear, right and left 2 meters from the vehicle. The whole ROI was partitioned into 3×3 grid regions, and the algorithm determined whether the parking line is crossed for each partitioned region by trained deep learning classifier, and finally a warning signal generated for the system. We also discussed the issue of porting to NVIDIA Jetson TX2 board to enable implementation of the parking line departure warning system into the vehicle.