3D Mapping of Rigid Body Feature Locations From Image Plane Measurements

This paper develops a method for determining physical locations of features on a rigid body from image plane measurements of a single camera over a period of time. The approach utilizes a total least squares algorithm, assuming the position and attitude of the rigid body relative to the camera are estimated externally or through a priori information. The method is useful for non-cooperative relative navigation cases in which there does not exist a CAD model or any other information regarding the dimensions and shape of the tracked object. Moreover, the method uses measurements from a single camera and is done without triangulation of multiple cameras. Such a method will be useful for on-the-fly autonomous tracking of non-cooperative objects, and for reverse engineering the physical shape and locations of prominent features relative to the body with quantifiable error bounds that achieve the Cramér-Rao lower-bound. The analysis shown in this paper thus achieves the best possible estimates of the features, and provides an assessment of the expected performance for any algorithms that requires physical feature information along with their associated error-covariances.