Automatic Lane Detection in Image Sequences for Vision-based Navigation Purposes

Intelligent Vehicles, as a main part of Intelligent Transportation Systems (ITS), will have great impact on transportation in near future. They would be able to understand their immediate environment and also communicate with other traffic participants such as other vehicles, infrastructures and traffic management centres. Intelligent vehicles could navigate autonomously in highway and urban scenarios using maps, GPS, video sensors and so on. To navigate autonomously or follow a road, intelligent vehicles need to detect lanes. It seems that the best cue for lane detection is to use the lane markings painted on roads and it should be noticed that among passive and active sensors, the video sensors are the best candidate for finding lane markings. In this paper we present a method for lane detection in image sequences of a camera mounted behind the windshield of a vehicle. The main idea is to find the features of the lane in consecutive frames which match a particular geometric model. The geometric model is a parabola, an approximation of a circular curve. By mapping this model in image space and calculation of gradient image using Sobel operator, the parameters of the lane can be calculated using a randomized Hough transform and a genetic algorithm. The proposed method is tested on different road images taken by a video camera from Ghazvin-Rasht road in Iran. Experimental results on different road scenes indicate the good performance of the proposed method.