Applying mathematics with a passion

LDW system investigation

ADVEA recently supervised RMIT University final year project student, Jake Werden, in investigating methods for developing lane departure warning systems virtually. The below summary of his work has been provided based on the final deliverable of his project.

In his project Jake proposes the testing of lane departure warning algorithms which can be used and integrated into majority of automobiles Australian roads today. The systems are to be tested through the Ground-truth to camera-detection algorithm (GTCDA) developed, using baseline models provided by Matlab and PreScan.

Jake mounted a GoPro camera in a vehicle and filmed driving a section of road, with lane changes, and then used Open Street Map to find a close representation to the practical path taken. The map exported used the the rectangle given by a longitude and latitude of (-37.72851, 144.90525) to (-37.73635, 144.88840) of the Tullamarine Freeway in Melbourne, Australia, as the segment of interest.

The Matlab Lane Departure Warning algorithm was modified for use on the new footage, and the algorithm was then applied to the PreScan-Simulink model used to simulate the environment.

Matlab LDW algorithm applied to the collected footage

A method was determined for comparing the ground truth (simulation) data to the actual collected data. The PreScan simulation was then used to test the same driving scenario with fog, sun glare, and the inclusion of solid lines rather than segmented lines.

Comparing ground truth simulation to the algorithm response to the collected footage

The results of the study indicate that environmental factors have a significant effect of the LDW systems when exposed to varying rain densities, fog, glare, different velocities and geometries. The LDW system preformed almost perfectly in optimal, light rain and sun glare scenarios. Fog, moderate and heavy rain scenarios caused a greater fluctuation in lane detection due to the LDW systems inability to define the lane markers correctly. When speeds were increased to 100km/h increased detection of lane changes however also increased the detection of false lane changes by 5 times. The real time, ground truth and PreScan camera detection methods preformed significantly well when drawing comparisons to their respective methods. Simulation and real-time portions of the algorithm not only detected the similar lane change areas within the segments of each scenario, yet also contained the same false lane detections due to divergence issues. The Ground truth to camera detection algorithm provided adequate results in allowing a combination of all methods to be advantageous and efficient.

No Comments Yet

Leave a Reply

Your email address will not be published. Required fields are marked *