Vision Tracking For Capturing Uncooperating Target Satellite Based On Kalman Filter

Catching target satellite could be used to refuel it, repair and save the failed satelliteand handle space debris. However, when the target satellite becomes invalid, it is verydifficult to catch defunct satellites. Tracking target satellites in time using vision-servo is anecessary premise for catching satellites. So this thesis is focused on vision tracking forcapturing uncooperating target satellite based on Kalman Filter. To simulate the realcondition of vision tracking of uncooperating target satellite in space, in this thesis,ADAMS software was used to generate predefined target’s motion to simulateuncooperating target satellite’ motion. According to the fact that uncooperating targetsatellite’s motion model is unknown, harsh light environment in space and some other realfactors, also the requirment of high-speed algorithm, this thesis chose Kalman Filter torealize vision tracking for targets. In addition, to increase tracking precision, Kalman Filterand its prediction model were improved under target’s three motion models includinguniformly linear, uniformly accelerating and un-regular curve. Since it is impossible togenerate predefined trajectory points (reference points) in camera’s coordinate systemdirectly, this thesis presented a mapping method to transform the points in the target’smoving plane to the ones in camera’s coordinate system without the help of optical localizer.To simplify the process of choosing Kalman Filter parameters and evaluating algorithm’sperformance, this thesis presented the other indicator besides the error covariance, which isnamed as the minimum square error (MSE) comparing the predicted points with therelevance points. In the experiments, it was found that the velocity in target’s state vectorconverged slowly and also the velocity component could not be measured directly, so inthis thesis, a velocity calculated by differencing positions was chosen to serve as themeasured velocity number to accelerate the speed of state vector’s convergence and thenincrease the tracking precision. Because the real motion model of the uncooperating targetsatellite is unknown, this thesis solved this problem by mixed model Kalman Filter. What’smore, to check the adaptability for noise of the algorithm presented in this thesis, Gaussiannoise and random noise were added to the relevance points separately and then thealgorithms which were improved were used to predict target’s trajectory in this condition.Through experiments’ analysis, the mapping algorithm presented in this thesis is right andpractical; the improvement of the algorithm and model’s optimization could increase vision tracking precision for uncooperating target satellite greatly; the improved Kalman Filter hasa better adaptability for noise comparing with the traditional Kalman Filter.