Recently, researches on swarm flight using a large number of drones are being actively conducted. To make a perfect formation of the swarms of drone flying so that they don’t crash into each other, precise relative position is absolutely important. To provide a number of drones with precise location and velocity information, emerging positioning techniques such as Real Time Kinematic (RTK) and Precise Point Positioning (PPP) are suggested and applied to several systems in a pilot manner. Even though both RTK and PPP are very suitable for providing cm-level high accurate position and velocity, there are critical limitations in computing positions of UAV, especially relative positions for swarm flights. When using RTK for calculating the relative position between UAVs, position accuracy and operating range are totally dependent on distance and geometry between the UAVs and ground GNSS reference stations (RSs). The RTK accuracy might not be still cm-level away from 20km or more from the station due to spatial decorrelation error of carrier phase measurements. What is worse, the reception rate of RTK correction that is transmitted by wireless communication channels would be far decreased when the drone moves away from the stations, and the position accuracy could be accordingly worse. When using PPP instead of RTK, 40 minute or longer initialization time for accuracy converging to cm level could be a problem due to the drone’s short operating time. Moving baseline RTK could be a proper solution to provide accurate relative position information for operating the drone swarm flight without limiting the mission area to near the ground station nor losing the operating time. Although a commercial service of moving baseline RTK had been used for landing on aircraft carrier, it is very difficult to use it in general because it can be exclusively used by customers who paid extra costs or because it does not disclose the necessary message protocol. To enable drones to use moving baseline RTK technique without any hardware nor geometrical constraint, we suggest a modification of conventional RTK to be suitable the dynamics of a leader and follower drones. To use the conventional RTK originally designed for a fixed RS to the system that both RS and rover are moving, we propose to compensate for the RS dynamic during the correction transmitting time. More specifically, the leader of the swarm flight is set to RS and followers to rovers, and the leader’s velocity vector should be sent to the followers with the RTK correction to find the relative position between the leader and the followers. Since the moving RS is not able to provide RTK correction message due to its built-in anti-spoofing function, we add a module to the leader UAV for encoding its raw measurement and rough location into RTCM v3 MT 1004 and 1003. The leader transmits its velocity vector calculated by time difference of carrier phase, and it enable the followers compensate for the leader’s movement during the correction transmit time. The followers’ construction has nothing different from the general RTK rovers, but it includes a latency compensation module to correct the calculated RTK baseline vector using the velocity vectors of the leader and followers. To estimate the correction transmitted time, correction age information in NMEA data is used. To assess feasibility our method in dynamic vehicle, we constructed two GNSS antenna with a fixed baseline on an automobile. While the conventional RTK whose RS is on the moving automobile computed its baseline vector with horizontal accuracy of 7.3m (RMS, Root Mean Square), our methodology can improve it up to 0.05m. We also get similar performance assessment after applying it to a leader and follower drones. Considering the high-rate control of the swarm flight, we computed 10Hz position output. Originally, correction message should be fed at the same rate as the position output, but 10Hz baseline vector can be obtained accurately without any latency thanks to velocity information and correction age in NMEA even if correction message is provided at 1Hz. The 10Hz horizontal baseline errors computed from the conventional RTK method without compensating the correction age were followed to a sawtooth-shaped trend of 1 second period and their RMS value was 1.4m. However, the baseline vector computed from our methodology could guarantee Gaussian-like horizontal errors of 0.1m accuracy (RMS) regardless of whether the correction message is received or not. We expect the suggested moving base technique would enable follower drones to compute high-rate accurate relative positions by being provided with only 1 Hz RTK correction and the leader’s velocity. Both RTCM for encoding/decoding the RTK correction and NMEA for computing correction age are well-known GNSS standards, thus this methodology can be generally applied to any type of receivers or formation flights without any extra charge or proprietary message information. Since the RTK correction message is transmitted from the near leader not a far ground station, higher receptance rate of the correction message and accordingly robust and accurate position results are also promising.