Spacecraft attitude estimation based on attitude-weighted kernel regression

An attitude estimation method for monocular vision imaging systems was proposed to solve relative attitude estimation problem of spacecrafts. On the basis of the original kernel regression model, the similarity of training data in the attitude space was used to weight kernel functions of the original kernel regression model in the visual input (image feature) space. A joint probability distribution function of input variables (image features) and target variables (attitudes) was learned, which was called acceptance function. Given images containing spacecrafts with unknown attitudes, the estimated attitude of the target spacecraft can be obtained by maximizing the acceptance function in the attitude space. The proposed method just needs training data to learn the model, so it has fewer limits than other vision based methods. Comparison experimental results show the advantage of the proposed method in attitude estimation. The effectiveness of the proposed method for spacecraft attitude estimation was also validated by the experimental results on satellite image dataset.