GPS traces can track actual time and coordinates of regular vehicles going their own business, and it is easy to scale to the entire area with an accuracy of 6 to 10 meters in normal condition. As a result, extracting road map from GPS traces could be an alternative way to traditional way of road map generation. The basic idea of this paper is to describe a process which incrementally improves existing road data with incoming new information in terms of GPS traces. In this way we consider the GPS traces as measurements which represent a "digitization" of the true road. Although the accuracy of the traces is not too high, due to the high number of measurements an improvement of the quality of the road information can be achieved. Thus, this paper presents a method for integrating GPS traces and an existing road map towards a more accurate, up-to-data and detailed road map. First we profile the existing road by a sequence of perpendicular profiles and get the road's candidate sampling traces which intersect with the profile. Then we match the potential traces with the road and finally estimate the new road centerline from its corresponding traces. In addition to the geometry of roads we also mine attribute information from GPS traces, such as number of lanes. Furthermore, we explore the benefit of an incremental acquisition by a temporal analysis of the data.