Modelling issues in vision based aircraft navigation during landing

This paper investigates the the feasibility of using visual and infrared imaging sensors to aid in the location of the aircraft position during operations such as landing in bad weather. The choice of the airport model used is crucial to algorithms which are used for position estimation based on pattern recognition. In this paper we describe the effects the choice of a model has on the behaviour of such matching algorithms. Three basic models are chosen: a line segment based model, an area based model and a texture based model. It is seen that a sparse line segment based model is not adequate to identify the runway since it matches a number of false artifacts in the image. An enhanced line segment based model containing a large number of features compares favourably with the area based model. The texture based model is seen to need a number of camera and weather dependent parameters and the performance of such a scheme is not seen to be substantially better. Thus either a proper area based model or a pseudo-area based model (based on a very large number of line features) can be seen to provide the best performance for such landmark identification and position determination algorithms.<>