Estimating pose and motion using bundle adjustment and digital elevation model constraints

The bundle adjustment is an algorithm for solving the simultaneous location and mapping problem by processing the information obtained from a sequence of images. As such, it suffers from the well-known ambiguity problem: Scaling cannot be recovered from visual data only. Several schemes have been proposed in the literature to circumvent this difficulty, mostly using additional sensing such as GPS or range measurements. In this work, a different route is followed: It is assumed that a digital elevation model is available for the region observed by the imaging system. Assuming that the approximate locations are known over the observation time period, it is shown that the bundle adjustment can be constrained by using the model to disambiguate the overall solution. Moreover, the new constraints do not cancel the desirable block properties of the algorithm, so that an efficient numerical algorithm can be readily obtained and requires minor modifications to the original unconstrained one. The properties of the new algorithm are first tested using a numerical study. Then the feasibility of using the algorithm is investigated by using data taken from a laboratory model and from an actual flight test.