Integrating Optic Flow, Ins and Gps Time Differenced Carrier Phase Measurements for Uav Surveying and Mapping

To improve the performance of UAV navigation and geo-referencing systems, this work has proposed an approach of integrating optic flow, INS and GPS time differenced phase measurements. Time-differenced carrier phase measurement (TDCP) is used to estimate the precise position changes (delta positions) of the GPS receiver between observation epochs, which then constitutes a precise aiding source of millimetre level accuracy for on-line INS error calibration. Optic flow has been widely accepted for implementing machine vision and automation. In a simplified model, the optic flow measurement is the sum of translational and rotational motions of the camera/host vehicle platform. After extraction of the platform’s angular movement, the remaining part is proportional to the ratio of the translational velocity and ground height. When the velocity measurement is available, the ground height could be estimated in terrain following tasks. Optic flow can also be used as backup aiding during GPS drop outs. With three types of measurements from GPS, INS and optic flow being optimally fused together, the integrated system represents significant improvement in robustness and reliability. A tightly-coupled complementary extended Kalman filter (EKF) is implemented for the proposed data fusion strategy. In this study, the proposed scheme has been implemented and evaluated with the field data collected from different platforms in a post processing mode. 1. INTRUDUCTION Having a robust and reliable navigation and geo-referencing system is of significant importance for flying a UAV. As for civil use, a combination of low grade Inertial Navigation System (INS) and Global Position System (GPS) is often the choice for providing the UAV navigation and geo-referencing information when considering the cost and payload limit. However, there are several known limitations with such low cost GPS/INS integrated systems, such as: 1) the positioning accuracy of single frequency standalone GPS receivers working in single point positioning (SPP) mode is normally much poorer than those achieved with differential GPS (DGPS), or real-time kinematic (RTK) GPS. In practice, the positioning accuracy of several meters is normally not enough for providing good calibration of low grade INS sensors before the inertial errors are built up. 2) such systems heavily rely on GPS signal availability. The system performance would degrade sharply without GPS aiding due to the poor performance of low cost INS. 3) UAV autonomously landing and terrain following operations need ground height information, which is not available from GPS/INS integration. Aiming to overcome these limitations mentioned above, an approach of integrating optic flow, INS and GPS time differenced phase measurements has been proposed in this research. The GPS carrier phase measurements could be chronologically differenced to derive an average Doppler measurement, which are then used for velocity estimation [Serrano et al., 2004a; Serrano et al., 2004b; A. Wieser, 2007; Andreas Wieser, 2007]. The results are more accurate than that obtained from raw Doppler measurements. This is because the carrier phase derived Dopplers are computed over a longer time span, the measurement noises are better suppressed. Also the large GPS error sources such as ionosphere and troposphere influences are significantly reduced due to their temporal correlation. Based on this, the time-differenced carrier phase measurements (TDCP) are used in this work to estimate the precise position changes (delta positions) between GPS observation epochs, which constitute a precise aiding source that is better than conventional velocity aiding. With time differenced carrier phase measurements from a single frequency standalone GPS receiver, the delta positions can be estimated at millimetre level accuracy[Ding and Wang, 2011; van Grass and Soloviev, 2004]. Optic flow is a biological inspired technology that has been widely accepted for implementing machine vision and automation [Barron et al., 1994; Garratt and Chahl, 2008; Horn and Schunk, 1981; Srinivasan, 1994; Zufferey and Floreano, 2006]. Since close objects exhibit a higher angular motion in the observer’s visual field than distant objects, the relative range between the feature objects in the environment can be determined by measuring the motions of visual features obtained from vision sensors. With a simplified model where the ground surface is assumed to be flat and static, the optic flow measurement is the sum of translational and rotations motion portions. After extraction of the angular movement, the host vehicle’s translational velocity can be estimated. When the three types of measurements, i.e. TDCP, optic flow and INS, are optimally fused together, the integrated system represents significant improvement of robustness and reliability in navigation and geo-referencing operations In this work, a tightly-coupled complementary extended Kalman filter is implemented for carrying out the data fusion. The dynamic matrix used for Kalman filter is based on INS psi error model[Da et al., 1996], with augmentation of optic flow error states. With the optic flow as additional type of observations in KF, it can be used as an effective measure to provide UAV ground height in terrain following tasks. Mean time, it also acts as a backup aiding to velocity estimation in case of GPS drop outs. This proposed scheme has been implemented and evaluated with the field data collected from different platforms in a post processing mode. In the following, measurement models of optic flow and time differenced carrier phase measurements are first introduced. Then dynamic error model used for the EKF is established based on the INS psi angle model and the perturbation of the optic flow measurement model. Then the optic flow and integration scheme has been evaluated in several scenarios. The tests are not conclusive, but appear to indicate that the new technique is promising. 2. OPTIC FLOW AND TDCP MEASUREMENTS 2.1 Optic Flow Measurements For the purpose of this investigation, it is assumed that a monocular vision sensor is mounted in a downward looking position on a UAV platform. The interior parameters of the vision sensor could be calibrated in advance in the laboratory environment. More importantly, the vision data need to be properly time synchronised with other sensors to a unified time scale, in our case, the GPS time. A simplified two dimensional optic flow model has been adopted in this work [Corke, 2004; Garratt and Chahl, 2008]. Assuming the ground surface is near flat and the UAV has no substantial vertical movement, with a down-looking video camera, the optic flow measurement is a sum of translational and rotations components. The model can be expressed as: y x x H V Q ω + = (1) x y y H V Q ω − = (2) where y x Q Q , = optic flow measurements y x V V , = translational velocities in camera frame y x ω ω , = camera rotational angular rates H = height from ground surface coordinate system 2.2 Time Differenced Carrier Phase Measurements The potentials of chronologically differenced carrier phase measurements in precision velocity estimation have been discussed in literature. [Kennedy, 2002; Serrano et al., 2004a; Serrano et al., 2004b; van Grass and Soloviev, 2004; Wendel and Trommer, 2004; Wendel et al., 2006]. The velocity estimation derived from time differenced carrier phases (TDCP) is more accurate than that obtained from raw Doppler measurements, and the carrier phase ambiguity has been cancelled in this process so that the high precision of carrier phase measurements can be really explored easily. Considering the specific UAV application scenarios, a method to precisely estimate the delta position vector was proposed by the author [Ding and Wang, 2011], and has been adopted in the work. As shown in Figure 1, P(t1) and P(t2) are the receiver coordinates at epoch t1 and t2, v(t1) and v(t2) are the instantaneous velocity vectors, and b(t1) the position change vector from epoch t1 to t2. The vector b(t1) is the baseline vector from point P(t1) and P(t2), which is called delta position here for convenience. The relationship between v(t1) and v(t2), and b(t1) is straight forward, which will not be detailed here. Although the difference between estimating instantaneous velocity and estimating delta position seems trivial, the later one has great advantages in the following filter design, which will be shown here. Figure 1. Calculation of delta position based on GPS carrier phase measurements 3. FILTER DESIGN 3.1 INS Error Model INS navigation solution drifts over time due to inherent bias of inertial sensors. Two approaches used for deriving INS error propagation models are well known in the field, namely the perturbation approach and the psi-angle approach. The psiangle error is the most widely adopted approach and has been implemented in many tightly coupled integration systems, as shown by Equation (3). ( ) ε ψ ω ψ g f ψ v ω ω v v r ω r + × − = ∇ + + × − × + − = + × − =