Vision-based Control in the Tracking of a Moving Ground Target for UAV

Abstract

The moving ground target’s position and motion cannot be measured directly in tracking it by unmanned air vehicle via camera vision. In this paper, a type of filter named continuous nonlinear Kalman filter is used to estimated target motion , directional heading on flat ground and distance from the UAV to target. Before the filter can be implemented, the kinematics of the tracking problem must be established, and then we obtained the nonlinear process model. Next, F(t) was computed based on
the assumption of the following constant values: Vg = 25/s, η= 0, range ρ= 400m, λ=Vg/ρ = 0.04 rad/s , Vt =5 m/s, ψg = 0.04 rad/s, ψg took on the latest value from the UAV truth model as the UAV changes heading. ψt was unknown and hence took on the value from latest estimated target heading ψt. Clearly, the gain varied according to the difference in heading
between UAV and target. This set of gains was used to provide estimates of the state variables. Equation was implemented using existing simplified UAV truth model based on only one body turn rate ψg in yaw, for the airframe. The nonlinear Kalman filter performance is shown. The error was
about 45 (after 500 seconds) compared to the true range of approximately 250m mean, meaning an error of about 18%. values assumed for variables in H(t) and the low values of the Kalman gain, for target velocity, which was related to the choice of noise covariance. Overall, the filter in this particular implementation could provide estimates of the state variables to within 20% error approximately except for the target velocity. In future, further assessment using Kalman filter technique will be beneficial in identifying the cause of the estimation discrepancies observed.