In this paper we have demonstrated a method for incremental improvement of existing road data with incoming, massive amounts of data possibly of low quality. Exploiting the potentially high amount of information compensates for the lower quality. We match new GPS traces with existing road information according to their distance to the road, direction and the angle between the trace and road. We use fuzzy c-means clustering method to separate traces when two roads are close and have similar direction. We also extract additional attribute information from GPS traces, such as number of lanes, turn restrictions of the roads. We plan to test the approach of extracting road centerline using data from urban area, where the situation is more complicated, and make some improvement to the approach if it is needed.