Lane Detection using Fuzzy C-Means Clustering

In general, road lane detection from a traffic surveillance camera is done by the analysis of geometric shapes of the road. Thus, Hough transform or B-snake technology is preferred to intelligent pattern matching or machine learning such as neural network. However, we insist that the feasibility of using intelligent technique in this area is quite undervalued. In this paper, we first divide the image into halves and use only the lower part in detection and binarize them by analyzing RGB channel. Then the boundary lines are extracted by applying 4-directional contour tracking algorithm and vectors with distance and angle values are extracted from those boundary lines to use as input for fuzzy C-means clustering algorithm. Clustered vectors form a straight line for road lanes by our method. In experiment, the proposed intelligent method is slightly slower than Hough transform but better in accuracy thus there is a room for intelligent method such as fuzzy C-means to solve this problem.