Multi-Lane Detection System using Perspective Transform and Curve Fitting for Autonomous Vehicles

This study presents a multi-lane detection system designed to accurately identify road lanes for autonomous vehicles. In the proposed system, perspective transformation and curve fitting techniques are applied to the images obtained from the vehicle's camera to detect lane markings. The lane detection results obtained from the YOLOPv2 model are further refined through contour detection and polynomial curve fitting processes, ensuring precise lane modeling. The system was tested on both straight and curved roads, achieving successful results in both scenarios. In future work, new tests and algorithm improvements will be conducted to enhance the performance under different road conditions and environmental factors. Additionally, future efforts will focus on detecting various lane types such as emergency lanes and side road connections.