Classifying self-cast shadow regions in aerial camera images

Abstract

In many fields of airborne surveillance, self-cast shadows (i.e. shadows in a scene which are cast by an aircraft) pose a not negligible problem for image processing tasks. Self-cast shadows can impede the stability of computer vision applications like remote sensing, visual odometry, or tracking tasks. In order to be able to reliably identify selfcast shadows in on-board camera images, a model-based approach has been developed. This approach utilizes easily accessible sensor data to make predictions about the position and the shape of self-cast shadows. The predicted shadow is then used to search for image regions which contain self-cast shadows. The developed approach is presented in detail in this paper. Further, the approach is applied on flight test data which has been recorded by a helicopter that is operated by the German Aerospace Center (DLR). The evaluation of the flight test shows that the approach is able to identify self-cast shadow regions with a high reliability.