Experimental Results on Edge-based Lane Detection

Abstract

This paper evaluated the experimental results of the proposed lane detection framework during heavy rain condition. The framework of the proposed lane detection system covered pre-processing stage and detection stage.  A pre-processing stage mainly consists of colour space conversion, selection of region of interest, edge extraction. Whereas, detection stage mainly consists of Hough transform and local maxima detection for lanes. The main contribution of this paper is to extend testing methodology for existing lane detection algorithms on stability, accuracy, and singularity. The experiment is conducted in urban roads of Malacca city. For experimental results assessment, coordinates variation analysis and uniqueness analysis are proposed to evaluate the performance of the proposed lane detection system, in term of stability, accuracy, and singularity. The test results are compared to Canny, Prewitt, Roberts, and Sobel methods for benchmarking. The experimental results shown that the proposed method has lesser fluctuation in coordinates variation analysis and closer coordinates mapping to their respective reference coordinates despite heavy raining weather. In addition, the proposed method has shown higher distinctive rate of correctly detecting the ground truth lanes despite certain frames are having outliers due to other uncertainty factors such as disturbances from wiper and rain drops on the front windscreen.