Sensor pose inference from airborne videos by decomposing homography estimates

Airborne videos are gaining increasing importance. Video cameras aretaking huge amounts of measurements for low costs. Their low weightand low requirement for energy makes them particularly attractivefor small airborne carriers with low payload. Such carriers are discussedfor military as well as for civil applications, e.g. traffic-surveillance.Often video cameras are used for documentation and reference in connectionwith other sensor systems. In addition to panchromatic or ordinarycolour videos, nowadays also cameras operating in the thermal spectraldomain gain attention. For the utilization of any stream of measurementstaken from a moving platform the pose of the sensor in orientationand position has to be constantly determined. For airborne platformsoften GPS and INS are used to acquire this information. However,the video stream itself provides also possibilities to estimate poseparameters. In this contribution we restrict our investigation toalmost flat scenes but we allow oblique views both forward lookingand side looking. The optical flow of the scene fixed structure onthe world plane is estimated by a planar projective homography. Thisrequires at least four point or line correspondences that can betraced over an appropriate number of frames. If the focal lengthis not changed and the camera has not been rotated, the proper transformwill be restricted to a planar homology with five degrees of freedom.For this case pose parameters were estimated by the analysis of theeigenvectors of the transform. The axis of the homology (correspondingto its double eigenvalue) is the horizon, from which we obtain therotational part of the pose. The sensor translation is obtained fromthe vertex (corresponding to its single eigenvalue) and the crossratio of the homology. Situations where the vertex is close to thehorizon needs special treatment (this mapping is called elation).If the camera has rotated between the frames the homography willbe decomposed into a homology and an orthogonal rotation matrix.The five degrees of freedom of the homology and the three degreesof freedom of the orthogonal rotation matrix sum up to eight, whichis exactly the same number of degrees of freedom that a planar homographyhas. There is a set of analytic solutions to this equation system,of which the correct solution can be picked by heuristic considerations.We also investigate the propagation of measurement errors throughthese calculations. Examples for such estimations are shown for athermal video.     «

Airborne videos are gaining increasing importance. Video cameras aretaking huge amounts of measurements for low costs. Their low weightand low requirement for energy makes them particularly attractivefor small airborne carriers with low payload. Such carriers are discussedfor military as well as for civil applications, e.g. traffic-surveillance.Often video cameras are used for documentation and reference in connectionwith other sensor systems. In addition to panchromatic or ordinarycolour videos...     »