On Precisely Determining Self-Cast Shadow Regions in Aerial Camera Images

This paper addresses the occurrence of self-cast shadows in on-board images of aerial vehicles which are caused by the Sun. Due to the shadows nature of modifying the observed scenery within these images, self-cast shadow poses a not negligible problem to several computer vision applications like remote sensing, visual odometry, or tracking tasks. Therefore, a possibility is needed to reliably identify self-cast shadow regions and to exclude them from further processing tasks. The proposed model-based approach achieves this by using data that is accessible for most aerial vehicles (i.e., data provided by an Inertial Navigation System and a geometrical model of the shadow casting object). In this paper, the algorithm to detect self-cast shadows in on-board images is presented in detail, focusing on its potential impact on visual odometry. This algorithm is applied to flight test data which has been recorded by an unmanned helicopter that is operated by the German Aerospace Center. The performance of the algorithm is evaluated by comparing the test results to empirically determined ground truth data. The results show an accuracy of close to 100 % in terms of finding the correct area of the self-cast shadow and a high similarity between the shape of the real self-cast shadow and the estimated shadow.