Real-Time Videogrammetry System for Navigation of Unmanned Aerial Vehicles

This proposal presents Cognitech's Phase II extensions on the theoretical and computational Phase I study of automatic landing of unmanned aerial vehicles (UAV). The Phase I report contained successful initial validation study with comparison of the platform positioning as estimated from visual and IR video sequences to the recorded GPS data. Our approach differs from other attempts to use images for automatic landing in that we utilize mainly detection and estimation of meaningful lines instead of reference points. The major advantage of our solution approach to the UAV automatic landing task is generality: no airport surveying data is necessary. We rely on only one natural parameter, the maximum false alarm rate, unlike all previous methods that depend on a multitude of unstable thresholds. We propose to make all computational tasks the real-time, in particular the detection of meaningful straight lines, and to assess the precision of the measurements with respect to the resolution of the camera. We also propose to develop an automatic camera calibration process that is easy to set up. Finally, the solution to an automatic video-estimation of the position of a hose in an aerial refueling system is developed.