Abstract: GNSS technique, represented by GPS, could take advantage of high positioning precision, safety and integrity to improve the performance of train positioning, which is crucial for the development requirements of train control system on safety, efficiency and economy. High precision track map is the foundation of GNSS based train positioning, and GPS based track map generation is one of the most important content of intellectualized train control. To meet different leveled requirements of train positioning and control, the track map data for train control consists of private data, track data and terrain environment data, and the track data are divided into feature layer and interpolation layer under different scales, which could contribute to complete structure of the track map. To huge GPS measurements, the multi-points weighted distance discrimination and Kalman estimation are applied to reject incorrect data; with improvement of Douglas-Peucker method, in which the original data are divided as the track curve feature and the limited error is adjusted dynamically, certain feature points are selected to describe the real track under low scale; based on the feature point set, interpolation points are generated by inverse calculation of cubic B-spline curve and equidistance interpolation as certain distance resolution, which could refine the track description under high scale. With the replenishment of mileage information, the generation of track map data could be realized completely. Calculation results with measurement data from Qinghai-Tibet railway show that, the proposed track map data generation method could realize the description and approximation to the real track under different scales efficiently with original GPS measurements, which could meet various requirements of train positioning and control, with simple calculation, engineering realization and high practical application value.