Development of a Real-Time Lane Departure Warning System for Varying Severity Levels Using a Sparse 3-D Lidar Sensor

The study proposes a mechanism to robustly identify lane markings using sparse point cloud and offers a practical solution for detecting different types of vehicle movement based on their severity levels which have not been addressed in existing literature. Ground surface was first separated out from 3-D point cloud data collected from lidar sensors in dynamic conditions using a random sample consensus (RANSAC) algorithm. Within the ground plane, curb points were segregated using a modified Z-score and a curb line was generated using RANSAC. After getting tilt information, outlier points were removed, and lane marking points were segregated from road surface based on intensity values. Further, a window search algorithm was employed to get the lane marking points within each window, and a ridge regression (regularization) method was used for polynomial fitting of lane marking lines. Based on the minimum vertical deviation, the best fitted lane line was selected and other lines were parallel fitted accordingly. The system was designed to continuously monitor the vehicle position compared with detected lanes. When a vehicle begins to drift within or outside its designated lane or approaches a lane boundary without the turn signal being activated, the system issues a warning to the driver. The proposed method achieves an accuracy of 99.87% and precision of 98.01%, demonstrating its robustness. With an average run duration of 40 ms per frame, the suggested algorithm can operate in real time. The proposed system will help drivers improve lane keeping behavior, maintain lane discipline, and prevent crashes resulting from unintentional lane departures.