Empirical Comparison of Face Verification Algorithms from UAVs

Face verification use cases have recently gained momentum in the increasingly digitalised society, and thus the need arises significantly to integrate this technology in wireless/mobile networked systems such as 5G and applications such as Unmanned Aerial Vehicle (UAV) based public safety services. However, there is no benchmarking result for the evaluation of the various existing face verification algorithms for such UAV applications. This paper is concerned with such new use cases (e.g., the Drone Guard Angel in the EU H2020 project ARCADIAN-IoT and the surveillance network applications in the EU H2020 project 5G-INDUCE), and provides an empirical comparison among three popular state-of-the-art face verification algorithms for this use case. To this end, a face verification pipeline is presented. These algorithms are then compared in terms of their inference time, and the distribution of the similarity indexes for different distances in UAV-based use cases. Their strengths and weaknesses are analysed, leading to an insightful recommendation on their applicability scenarios for UAVs.