Abstract: To solve the problem that conventional obstacle identification methods cannot be adopted for real-time and precise obstacle identification during girder welding, obstacle detection of complicated seam track based on RNGK-FCM was put forward. Real-time clustering strategy was introduced, distance function was replaced, and consequently global fast optimization was realized. Through the comparison and analysis of the clustering performances of FCM algorithm on MATLAB platform, it was proved that RNGK-FCM obstacle detection method achieved the real-time clustering strategy to gain large number of clusters, and had better robustness to ignition noise. The sensitiveness to initial value was reduced and the high accuracy of clusters identification was boasted compared to traditional FCM. A real-time obstacles recognition experiment was carried out in the automatic grinder welding production line of a certain company and it was accurate in clusters calculation, good in real time and precise in obstacles avoidance, providing a strong foundation for automatic welding of girder.