Edge-based Realtime Image Object Detection for UAV Missions

Unmanned Aerial Vehicle (UAV) has limited computing power, but requires high accuracy and low latency in the visual object detection for critical UAV missions, such as infrastructure inspection. It may need highly complex machine learning algorithms with the demand of extensive computing power. With the rising edge computing technology, the heavily-loaded object detection tasks can be offloaded to edge computing systems. To enable such edge-based object detection with low overhead, we discover that it is critical to minimize the response time of the detection while maximizing the frequency of detected image frames. In this paper, we identify three key research challenges, conduct an experimental case study to show that current edge-based naive solutions cannot achieve the above goal, and finally point out major ideas for potential solutions.