Multi-UAV Sensing Over Urban Areas via Layered Data Fusion

In this paper, we address the problem of layered data fusion occurring within a collaborative and distributed sensor network. Multi-UAV sensing combined with stationary sensors has the potential of creating smart airborne systems for completing aerial tracking and surveillance missions over an urban area. However, matching two different data with multiple resolution reveals the presence of large motion and dominant outliers, which makes this problem very difficult. Here we extend our earlier work to this type of multiple mini-UAV sensing by applying an information-theoretic cost criterion and cooperative optimization method. Assuming no zooming in the same type of cameras, we treat changes in scale-space as changes in UAVs´ altitudes, and equivalently as changes in data resolution. Scale-space analyses indicate the importance of a prior adjustment of uncertain scaling factors. That is, to achieve a better performance in layered data fusion, it is critical to apply an approximate scaling factor first, within some error bounds, before applying a robust data alignment algorithm directly. We apply our layered data fusion technique to several examples: a single video registration, a video registration with a reference image, and the alignment of two video sequences. Experimental results show that our method can be useful for aerial monitoring and tracking systems within a dynamic sensor network.