Multistage Vision System for Road Lane Markings and Obstacle Detection

This paper is to introduce a multistage vision system to detect relevant information for road driving. The system applies fuzzy theory in order to handle less information per image that classical methods use to do. The image processing is based on the fact that the image is splitted up into six horizontal bands and consists of three basic stages: local, regional and global analysis. Firstly, pattern matching is applied to the information obtained from edge detection of the image in order to determine local appearances of lines on the road. Secondly, and with the aim of increasing the final reliability on data, regional analysis (at the level of a band) of the coherence of the local appearances is performed. Thirdly, results of previous stages are globally analysed, making up a fuzzy description of the image from the information given at each band. In the global analysis, the information of the image being processed is combined by means of a fuzzy algorithm, with the information detected in previous images in order to conclude a final description of the road lane markings.