LANE LINES DETECTION ALGORITHM FOR SELF-DRIVING CARS

الملخص

A major reason for many traffic accidents is due to driver distraction and speeding. To assist drivers in their daily routine of operating an automobile, there has been significant research into developing Advanced Driver Assistance Systems (ADAS). ADAS systems will be able to observe and react to their surroundings so they can help prevent collisions from occurring and/or aid in navigating through difficult situations. One of the most important components of ADAS systems is the ability to recognize lanes on a roadway. Recognizing lanes is key to lane departure warnings, autonomous vehicle navigation, and collision avoidance.

The purpose of this paper is to present a robust lane detection algorithm for detecting lane lines that utilize an optimized version of the Canny edge detection process and the Hough Transform to accurately identify the location of lane boundaries in various types of environmental conditions. The proposed methodology consists of three main phases: pre-processing, feature extraction, and lane recognition. In the pre-processing phase, the raw input roadway image taken by a front facing camera is first converted to grayscale, and then processed using a series of Gaussian and Sobel filters to extract edge characteristics from the image. The Canny edge detection process is utilized to isolate the edges associated with the lane markings, and then a Region of Interest (ROI) is defined in order to limit the amount of processing time required to identify lane markings by eliminating unnecessary parts of the image, including the sky and roadside objects. In the final phase, the Hough Transform is used inside the ROI to locate and map the detected lane markings.