DroneCOCoNet: Learning-based edge computation offloading and control networking for drone video analytics

Multi-Unmanned Aerial Vehicle (UAV) systems with high-resolution cameras have been found useful for operations such as smart city and disaster management. These systems feature Flying Ad-Hoc Networks (FANETs) that connect the computation edge with UAVs and a Ground Control Station (GCS) through air-to-ground wireless network links. Leveraging the edge/fog computation resources e ectively with energy-latency-awareness, and handling intermittent failures of FANETs are the major challenges in supporting video processing applications. In this paper, we propose a novel “DroneCOCoNet” framework for drone video analytics that coordinates intelligent processing of large video datasets using edge computation o oading and performs network protocol selection based on resource-awareness. We present two edge computation offloading approaches, i.e., heuristic-based and reinforcement learning-based approaches. These approaches provide intelligent task sharing and co-ordination for dynamic o oading decisionmaking among UAVs. Our scheme handles the problem of computation o oading tasks in two separate ways: (i) heuristic decision-making process, and (ii) Markov decision process; wherein we aim to minimize the total computation costs as well as latency in the edge/fog resources while minimizing video processing times to meet application requirements. Our experimental results show that our heuristic-based o oading decision-making scheme enables lower scheduling time and energy consumption for low drone-to-ground server ratios. In comparison, our dynamic reinforcement learning-based decision-making approach increases the accuracy and saves overall time periodically. Notably, these results also hold in various other multi-UAV scenarios involving largely di erent numbers of detected objects in e.g., smart farming, transportation tra c flow monitoring and disaster response.