The accurate distance and speed estimates of train and individual coaches are necessary for the safe operation of the high-speed train system. Often, the train system does not rely on a single sensor for its distance and speed measurements as the sensors are susceptible to diverse operating conditions such as snow, rain, fog, tunnel, hilly region, slip, slide, etc. Hence, the information from a combination of sensors which can complement each other under certain operating conditions is required for the correct estimation. For distance measurements, Global Navigation Satellite System (GNSS) and balise are generally used. For speed sensing, the combination of wheel sensor, radar and GNSS are chosen. The diversity of sensors in terms of sampling rate and noise characteristics, etc. greatly affect the overall estimation accuracy and reliability if the measurements are used directly. Hence, this work presents a probabilistic weighted fusion algorithm which is based on the nonlinear longitudinal train dynamic model. The fusion algorithm combines the state estimates from distributed and sensor-specific extended Kalman filters. The effectiveness of the proposed fusion algorithm is demonstrated on the simulated sensor measurements along with a wide range of noises, spurious measurements, train operating conditions and track environmental disturbances.