DeepBrain: Experimental Evaluation of Cloud-Based Computation Offloading and Edge Computing in the Internet-of-Drones...

Area of Science:

• Computer Science
• Electrical Engineering
• Aerospace Engineering

Background:

• Unmanned Aerial Vehicles (UAVs) are crucial for Internet-of-Things (IoT) and smart city applications, collecting vital aerial data.
• Real-time data processing on edge devices for drones (Internet-of-Drones) is challenging due to limited UAV energy and resource-intensive computer vision, especially deep learning algorithms like Convolutional Neural Networks (CNNs).

Purpose of the Study:

• To propose a system architecture for computation offloading in Internet-connected drones.
• To experimentally evaluate and compare the performance of cloud computation offloading versus edge computing for deep learning applications on UAVs, focusing on energy, bandwidth, and delay.

Main Methods:

• Developed a system architecture for computation offloading in Internet-connected drones.
• Conducted comprehensive experiments to assess energy consumption, bandwidth usage, and communication delays.
• Investigated the trade-off between communication costs and computation for cloud offloading and edge computing approaches.

Main Results:

• Computation offloading achieved significantly higher throughput (frames per second) compared to edge computing.
• The study experimentally validated the performance trade-offs between communication costs and computation for both approaches.
• While cloud offloading introduced larger communication delays, it provided superior processing speed for deep learning tasks on UAVs.

Conclusions:

• Computation offloading is a viable strategy to enhance the performance of deep learning applications on UAVs, overcoming the limitations of edge computing.
• The findings highlight the critical trade-off between communication latency and computational efficiency in drone-based data processing.
• This research provides valuable insights for optimizing resource allocation and system design in the Internet-of-Drones ecosystem.