This paper introduces an innovative automatic road‐vectorisation algorithm based on dynamic pixel clustering using particle swarm optimisation. A new cost function is designed to optimise the number and position of road keypoints and is capable of deriving road centrelines without considering geometric, spectral or topological characteristics in the road model. The algorithm is applied to different high‐resolution images (IKONOS, QuickBird and aerial photographs) and is evaluated with respect to RMSE, correctness and completeness. Moreover, a new quality parameter is defined to evaluate a “kinking” effect in roads. Extraction of different road shapes with an acceptable precision in both urban and rural environments proves the efficiency of the algorithm in yielding complete road networks.

No abstract

This paper introduces a novel road extraction algorithm in two stages of road detection and road vectorization. In the road detection stage, road class image is obtained using fuzzy C-means clustering and some post processing operations. In the vectorization stage road key points on the road centerline is obtained by an innovative approach of dynamic road pixels clustering using particle swarm optimization. The proposed algorithm is able to automatically optimize number and position of road key points without considering the prior information about the initial number and position of cluster centers by designing a new cost function. The optimized road key points were connected using weighted graph theory. Different high resolution images of Ikonos in urban, non-urban, and mountainous areas were tested and several quality measures including RMSE, correctness, completeness, and quality were calculated. Extracting different road shapes with RMSE less than 1.3 and quality greater than 0.86 in different areas proves the efficiency of the algorithm in yielding complete road networks.