LaneDetect: An Efficient and Reliable Lane Detection Method using Perspective Transformation and Polynomial Fitting

A significant development is underway in autonomous vehicles and driver assistance systems, which relies on effective lane detection for safe roadway travel. This paper presents a lane detector based on sobel filtering to obtain effective edge detection and polynomial fitting for reliable lane modeling. The lane detector is able to detect lane markings with high reliability even in challenging scenarios such as low-light conditions, shadowy patches on the roadway, and roadways with drastic curvature. A sliding window approach is also adopted to enhance lane detection further, especially as it relates to tracking the lane pixels that dynamically vary with the changing road structure. We then systematically enhance lane detection through polynomial regression and ensure smooth, coherent lane tracking, which addresses the issues confronting traditional edge-detection algorithms in the presence of noise and extraneous materials. The proposed lane tracker is also computationally feasible, and thus practical to deploy into real-time driver assistance applications such as ADAS (advanced driver assistance systems) and self-driving vehicles. We thoroughly present an experimental demonstration on varying roadway conditions and will demonstrate the viability of the lane tracker to cope with most variances to enable future development for intelligent transportation systems implementing roadway lane tracking.