Designing a Machine Learning Based Efficient and Outlier Free Lane Detection System for Autonomous Vehicles

The high growth in the number of automobiles has raised the demand of enhanced driver and passenger safety measures. The goal being effective and efficient application of evolving technology to address the concerns arising with partially or fully autonomous vehicles. Accurate lane positioning and lane change detection is an integral part of Advanced Driver Assistance System which helps in safer trajectory planning. The engineering marvel of Light Detection and Ranging sensors corrects global localization error and provides centimeter-level accuracy but finds its use in restricted domain due to its computational complexity and cost implied. Lane detection is a technique for automatically road markers to ensure that cars stay in their assigned lane and do not collide with vehicles in other lanes. As an alternative, machine learning enabled vision-based lane change detection is highly regarded to provide lane-level localization maintaining high level accuracy. The algorithm design provides polynomial fitting with LiDAR data that helps model perform well in bad lighting conditions and help detect curved lane marking. The technology is well adopted based on the real-world testing scenarios which affirms its real time robustness with average of 89% lane change detection accuracy.