Performance prediction and selection of aerial perception functions during UAV missions

The increasing automation of unmanned aerial vehicles (UAV) for versatile mission scenarios claims the challenge of implementing sophisticated perception functions. Thereby, environmental conditions (e.g. weather conditions or landscape changes) affect the performance of such functions for navigational (e.g. sense & avoid) or surveillance (e.g. object detection or tracking) purposes. This paper proposes an algorithm selection method to maintain the overall performance for surveillance missions. The algorithm selection method uses fuzzy inference systems (FIS) trained on example data to predict the perception functions performance under different environmental conditions. The experimental setup comprises four different ground vehicle detection algorithms evaluated on an example observation mission. The results show that the method is able to estimate the algorithms performance and increases the overall performance over mission time compared with using only one algorithm.